The Peter Attia Drive - #363 ‒ A new frontier in neurosurgery： restoring brain function with brain-computer interfaces, advancing glioblastoma care, and new hope for devastating brain diseases ｜ Edward Chang, M.D.

00:00:00.000 Hey, everyone. Welcome to the Drive podcast. I'm your host, Peter Atiyah. This podcast,

00:00:16.540 my website, and my weekly newsletter all focus on the goal of translating the science of longevity

00:00:21.520 into something accessible for everyone. Our goal is to provide the best content in health and

00:00:26.720 wellness, and we've established a great team of analysts to make this happen. It is extremely

00:00:31.660 important to me to provide all of this content without relying on paid ads. To do this, our work

00:00:36.960 is made entirely possible by our members, and in return, we offer exclusive member-only content

00:00:42.700 and benefits above and beyond what is available for free. If you want to take your knowledge of

00:00:47.940 this space to the next level, it's our goal to ensure members get back much more than the price

00:00:53.200 of a subscription. If you want to learn more about the benefits of our premium membership,

00:00:58.000 head over to peteratiyahmd.com forward slash subscribe. My guest this week is Dr. Edward

00:01:05.980 Chang. Edward is the chair of neurosurgery at UCSF and a leading innovator in functional neurosurgery

00:01:12.280 and brain-computer interface. Edward's work bridges the operating room, the research lab,

00:01:17.580 and the engineering bench to restore speech and movement for patients who have lost these traits.

00:01:23.200 In this episode, we discuss how modern neurosurgery evolved, dramatically reducing collateral damage

00:01:28.560 and recovery time, what happens during awake brain surgery, why the brain feels no pain,

00:01:34.060 how real-time mapping protects language and motor function, and the split-second decisions surgeons

00:01:38.800 make at the edge of the eloquent cortex, breakthroughs in brain-computer interfaces,

00:01:44.480 neural engineering's next frontier fully implantable wireless brain-computer interfaces,

00:01:49.840 and functional electrical stimulation systems that may bypass damaged nerves to restore breathing or

00:01:56.540 limb control, how genomic profiling, immune-based strategies, and more extensive resections are

00:02:02.440 slowly turning glioblastoma, a once uniformly fatal tumor, into a slightly longer survivable disease,

00:02:10.480 Edward's vision for 2030 and beyond, slimmer, safer brain implants to restore speech for people

00:02:16.380 with paralysis and other injuries, and how advances will help turn conditions like ALS,

00:02:22.760 spinal cord injury, and even aggressive brain tumors into more chronic, manageable illnesses.

00:02:28.680 So, without further delay, please enjoy my conversation with Dr. Edward Chang.

00:02:38.520 Eddie, thank you so much for taking a time out of your very busy schedule to come to Austin.

00:02:43.000 Really excited to talk with you today.

00:02:44.160 Oh, I'm thrilled to be here. Thanks, Peter.

00:02:45.680 Yeah. So, there's so much I want to talk about with respect to what your career is about today,

00:02:50.540 and what the field of neurosurgery is in today, and how the bounds are really being pushed. But

00:02:55.960 as we were talking earlier, I think that neurosurgery remains a little bit of a black box,

00:03:00.960 and it might help orient our listeners if we give a little bit of a history lesson. So,

00:03:07.880 can we orient ourselves back into the latter part of the 19th century,

00:03:13.360 and what were the typical problems that would have presented to a neurosurgeon,

00:03:18.220 and what were the tools that they had at their disposal? And let's posit that we're speaking

00:03:21.560 after the development of anesthesia, at least, so we're not in completely gruesome lands of

00:03:26.340 holding people down.

00:03:27.500 That is a really interesting question. And one of the reasons neurosurgery is a little bit of black 1.00

00:03:31.900 box is, in many ways, people consider it sort of like extreme medicine. It's like a very small

00:03:37.520 group of physicians that are taking care of patients with fairly severe indications,

00:03:42.760 a really rarefied field that takes a very long training in addition.

00:03:46.180 But let's say we go back 100 years, we're talking about the era of Harvey Cushing,

00:03:51.560 who's considered really the father of modern neurosurgery. I think that was a clear inflection

00:03:57.760 point in the history of medicine, in the history of neuroscience, in the history of neurosurgery,

00:04:02.500 really the beginning of what we'd call the modern neurosurgery. Why I think that Cushing was so

00:04:08.300 powerful was his observation, in addition to his ability to do extraordinary surgeries. So in addition

00:04:17.080 to being really an astute observer, in addition to being an incredibly technically skilled surgeon,

00:04:24.260 I think he was also an incredible internist too, diagnosing some of the first pituitary tumors and

00:04:29.840 effects of those on endocrine function. And then really the era of modern tools of craniotomy,

00:04:36.760 opening the skull to get access to brain tumors. And everything followed since then. The main

00:04:43.640 categories of neurosurgery have to do with tumors, the vascular system, which are aneurysms and strokes

00:04:50.080 and blood clot, spine. And then probably the most recent one is the one we call functional,

00:04:55.460 which actually has to do with understanding the functions of brain circuit, but also intervening to

00:05:00.920 change how they work using deep brain stimulation or other ablation methods. And those are the really

00:05:06.600 exciting new developments. And I think Harvey Cushing would be credited for the development of the

00:05:11.500 electrocautery as well, wouldn't he? Absolutely. It's hard to imagine that you could operate without

00:05:15.940 one of those things. Yeah. And not just in the brain, but anywhere in the brain. Yeah. It's like the key

00:05:20.620 of controlling bleeding and any surgery, but particularly in the brain, it's a tricky thing. And so Harvey

00:05:26.940 Cushing was just the starting point of modern neurosurgery. Then there's Wilder Penfield,

00:05:31.920 who was an American, but really did some incredible work in creating the Montreal Neurological Institute.

00:05:39.380 And that was really the beginning of what we'd consider modern epilepsy surgery. So surgeries

00:05:44.300 that are designed to stop people from having seizures. And he popularized this thing that we

00:05:51.140 all learn in medical school called the homunculus. It's this picture, right, of the little man there and

00:05:57.840 essentially the part of our brain that controls every muscle in our body and how it's laid out in

00:06:02.780 that particular part of the brain. It's something that we all learn. He was also a brilliant scientist

00:06:07.840 who helped us understand some of the basic things that we know about language. And from a technical

00:06:13.560 perspective, really popularized and developed the concept of awake brain surgery. And that's really

00:06:20.960 something that's captivated me since medical school and what I now specialize in now.

00:06:25.300 So let's fast forward a little bit into an era before you and I were in medical school,

00:06:32.840 call it the 70s and the 80s. What was the state of the art, call it 40, 50 years ago,

00:06:37.920 with respect to the vascular management, the oncologic management of masses in the brain

00:06:44.640 relative to today? So exclusive of the interventional side of things, just in terms of

00:06:49.800 being able to operate on the brain, where were the plateaus in technology?

00:06:53.240 What's interesting about it is some of the things that we do now are almost identical to the way that

00:07:01.160 Cushing did it over 100 years ago. And then some of it is radically different. So one of the work

00:07:08.180 course surgeries that we do is called a craniotomy. And that basically means where you remove a piece

00:07:13.880 of bone temporarily, you place it at the end of the procedure to access something like a brain tumor

00:07:19.080 that's in the frontal or temporal lobe. That's still being performed today and still really

00:07:23.600 indicated. But where we are now, there's ways of using laser probes through very small incisions to

00:07:29.860 get to deep targets in the brain to ablate them. There are ways of now even using focus ultrasound

00:07:36.200 that can be targeted to specific nuclei in deep parts of the brain in order to control someone's

00:07:43.080 tremor. For example, this is what we would consider a relatively non-invasive approach to do a neurosurgery.

00:07:49.820 So things have changed radically. I would say one other area where we have seen tremendous

00:07:56.060 disruption actually is in the vascular neurosurgery field. So back in the 80s and 90s, you'd have a large

00:08:03.340 craniotomy and incision, probably about seven to nine inches long, removal of a piece of bone,

00:08:09.280 and then using an operative microscope to dissect down to the deepest parts where the blood vessels

00:08:15.360 are coming to fix, let's say, an aneurysm, which is essentially a ballooning of a blood vessel that

00:08:20.600 when it ruptures can be fatal. And so that is a lot of what I trained on. Nowadays, 90% of those

00:08:27.960 procedures are now done through a catheter in the groin that's visualized. We put coils into the

00:08:35.260 aneurysms to help secure them. We can now do stents. Huge change actually now has been able to,

00:08:41.780 now being able to retrieve and dissolve clots that are causing acute, severe strokes. Those are

00:08:48.720 probably the most dramatic things that we've seen. Someone come in not able to talk, paralyzed on

00:08:54.360 half side of the body. In the old days, we would just give some medications, keeping our fingers

00:08:59.920 crossed that that would work. It rarely worked. It was really the minority of cases that it helped.

00:09:05.720 But nowadays, there's a much bigger fraction of patients where you can put the device in,

00:09:09.900 retrieve the stroke. And these are really game-changing things when someone can go home

00:09:14.000 the next day after such a huge thing. So we're starting to see things like stroke become more

00:09:18.800 like heart attacks. The cath labs are not just treating heart attacks, but they're also treating

00:09:23.500 what we call brain attacks or strokes. So some things are relatively similar to 100 years ago,

00:09:30.080 and then some things have just been totally changed.

00:09:33.500 So if 30 years ago, a neurosurgeon was doing a cratiotomy, virtually every time they were

00:09:40.060 addressing a pathology, today it might be less than half the time?

00:09:44.020 Yeah, that's right.

00:09:45.180 That's not a dissimilar parallel to what we see in other parts of surgery. I was just talking to

00:09:50.240 a vascular surgeon a couple of weeks ago, finished his training around the time I did. And I said,

00:09:54.680 you know, how many open surgical procedures are you doing? You know, I was talking about triple A's 1.00

00:09:59.340 and all sorts of, you know, fem pops and things like that. And he said, yeah, we do very little

00:10:04.040 open these days. It's virtually all done with stents, which again, even see like carotid arteries,

00:10:10.700 the whole thing. I was really blown away at how little they are operating now, meaning operating in

00:10:15.600 an open sense. Obviously they're still intervening.

00:10:17.360 Yeah, absolutely. I think what's happened with surgery is that this is not a trend. This is the

00:10:22.720 force of evolution that is guiding us towards things that are more minimally invasive, less

00:10:28.060 collateral damage to get to the targets and getting people back to life sooner. And it's a very exciting

00:10:34.860 time actually to be in medicine with all this technology that's coming on board. If you're okay

00:10:39.980 with that change, it's absolutely thrilling time. Yeah.

00:10:42.300 Now, I know this is not your field of interest. So feel free to say, yeah, I don't know enough

00:10:47.360 about it. But whenever I think of the brain, I think of GBMs. I think of these awful tumors, 0.97

00:10:52.660 which I guess for the listener, we can explain what that is. So maybe tell folks what a GBM is,

00:10:56.980 why a GBM is truly one of the cancers that gives cancer a bad name. And I guess my real question for

00:11:03.260 you is given that traditional surgery has historically never worked for this tumor, you simply can't resolve

00:11:09.100 it. And you think about the transition away from craniotomy, big open procedure. Is there anything

00:11:16.000 on the horizon for GBMs to render them less lethal? Yeah. So GBM stands for glioblastoma

00:11:23.120 multiforme, and it's a mouthful of a word. If you break it down, what it's referring to,

00:11:30.640 the glial part of that word refers to the cell origin. The brain has different types of cells. And if you

00:11:36.260 look in the big buckets, there's neurons, which are the ones that primarily are the ones that allow

00:11:40.620 us to do thinking and function. And then there's a large population of support cells that we call

00:11:46.220 glia. And that's that first part. Glioblastoma originates from those support cell population.

00:11:53.780 Multiforme refers to these original descriptions histologically of the tumors that really showed

00:12:00.360 multiple form, multiple histology. Some of the key features of it are the necrosis,

00:12:05.780 the tumor grows so quickly that it outstrips its blood supply. And in its wake, it leaves

00:12:10.760 cell death, what we call necrosis. So like you alluded to, it really is a terrible disease and

00:12:17.560 condition. We are making progress, and specifically around understanding what causes. So just 10 years

00:12:25.300 ago, you would remove one of these tumors, you send it to the lab, and you could get the diagnosis

00:12:30.300 that this is a glioblastoma. Now, in most academic medical centers, you'll also get a genetic profile

00:12:37.680 of the tumor. So nowadays, we actually know specifically what kind of mutations are involved

00:12:44.080 in the tumor. And that's going to be really critical for this next chapter, which is

00:12:48.020 using those genetic alterations, actually, to tailor and personalize chemotherapy and more.

00:12:54.140 So this has big implications because we're now moving from an era where we use a visualization

00:13:00.140 of the histology now to this molecular profiling, which is more mechanistic. The new chemo agents are

00:13:06.380 really going to be targeting mechanisms as opposed to general things like cell cycle and metabolism and

00:13:12.140 things like that. So these are the things that are changing, and we need it to change faster.

00:13:18.720 There are also really exciting things that we're seeing around new ways to train the immune cells,

00:13:22.960 to target things. It turns out that glioblastomas actually suppress parts of the immune system. So

00:13:28.780 they're kind of like growing in stealth, and they activate molecules and cells in a cloak way that

00:13:34.800 can't be recognized by immune cells anymore. And so if we can basically allow the tumors to be

00:13:40.340 recognized by the immune system, that could be something that really unlocks therapy in the future too.

00:13:44.960 But the things that we do know that work pretty well right now, at least in terms of prolonging

00:13:50.960 survival for patients and really meaningful survival, actually still is around the surgery. We do know

00:13:56.920 that the more extensive the resection, the longer the survival is, and that's been really well

00:14:01.860 characterized now. But it's not curative, like you said. We can remove 99% or even 100 and beyond

00:14:08.280 what we see on the MRI. Unfortunately, there's usually microscopic cells that go beyond what we can see

00:14:15.080 on the MRI that are still there and over time will repopulate the tumor. It is a really complicated

00:14:22.420 and tough disease, but we're working really hard on it. Do we have any idea what predisposes an

00:14:27.540 individual to this from a risk perspective? Short answer is no. Because it afflicts young,

00:14:31.860 it afflicts old. I mean, I've watched children die of this, teenagers, people in midlife, people at the

00:14:37.260 end of life. It seems to have no apparent pattern. Yeah. And part of that has to do with its mechanisms.

00:14:43.180 It's not something that we consider as a heritable risk, but what it does rely on is a set of

00:14:49.940 mutations, and it's rarely the same set. Right. It's a very polygenic condition.

00:14:54.680 Exactly. And that's what makes it really tricky to treat. When we talk about glioblastomal,

00:14:59.480 we're actually not talking about one thing. We're talking about a system of genetic alterations that

00:15:06.600 together have cascaded into the form that we see. And I suspect we'll come back to this,

00:15:12.180 Eddie, but you've alluded to chemotherapeutic agents that can be used, whether it's to treat

00:15:17.120 glioblastomia or anything else, including METs of other epithelial cancers that spread to the brain.

00:15:22.860 The blood-brain barrier poses a challenge for treatment. Do you think the future lies in

00:15:28.840 treatment within the CSF? So treating directly inside, intrathecally or directly into the central

00:15:36.360 nervous system? Or do you think it's designing drugs that cross the blood-brain barrier? I mean,

00:15:40.480 what do you think the future looks like? I think it's going to look like all of the above.

00:15:43.860 This is a situation where we do need to look at all possible options. This is not like the kind

00:15:49.100 of thing where we're thinking like non-invasive or minimally invasive, really something that will

00:15:54.140 work as the first priority. One of the technologies I'm really interested in following how this develops,

00:16:00.360 and we're doing research on this at UCSF, is using focus ultrasound. So most people know about

00:16:06.000 ultrasound to diagnose, but if you change the energy profile of it, you can actually use acoustic

00:16:11.800 energy through an ultrasound to actually open up the blood-brain barrier in targeted parts of the

00:16:17.800 brain. And so there is a lot of development on using that as a way to do delivery as opposed to

00:16:23.840 putting a catheter or something directly in the brain. And then with that set of new agents that can

00:16:28.960 be really molecularly specific to get to targets once you open up that blood-brain barrier.

00:16:34.380 What attracted you to neurosurgery? Was it something you knew you wanted to do when you

00:16:38.220 went to medical school or did you figure it out well there?

00:16:41.000 I had a sense, it was probably latent. I always knew that I was really interested in neuroscience,

00:16:46.480 the general field. It wasn't until I was in medical school that I was actually exposed to it.

00:16:53.180 And I remember really clearly in my first year, I had this neuroanatomy professor, Diane Rawson,

00:16:59.760 who was really incredibly kind person who was patient with me as we were learning. You probably

00:17:06.100 remember in med school, like learning hundreds of different parts of the brain anatomy.

00:17:11.460 All of which I've forgotten. I know there's a brain stem somewhere in there.

00:17:14.620 There's somewhere, something like that. Yeah. So that's part of our ritual, right? In medical school

00:17:18.940 to learn all of those terms and locations. But as you know, certain teachers just make such a huge

00:17:24.900 difference. She took me to the operating room one day and I saw one of my mentors. At the time,

00:17:31.200 I was just a student, but he ultimately became my mentor, Dr. Berger, doing awake surgery on a patient

00:17:36.640 with a glioblastoma. The surgery itself, I thought, was pretty interesting. But the part that left me

00:17:43.820 awestruck and the part that basically made it very hard for me to sleep for a couple days

00:17:48.200 was really just seeing an exposed brain, what the cortex looks like. The cortex is the outermost part

00:17:56.140 of the brain. It pulsates, it moves, but those are not from the mechanics of the brain itself. Those are

00:18:02.420 all just from the breathing and the heart rate, etc. The blood flow is coming through. So the thing that

00:18:09.080 really struck off for me was seeing a patient talking and not fully comprehending, but really being in awe of

00:18:16.240 the computations that must be happening in this part of the brain. And it's not like you're looking

00:18:20.940 at a computer. You're looking at essentially an organ composed of biological cells, 86 billion to be

00:18:28.420 precise, you know, how many neurons there are in the human brain. So that scene to me was deeply inspiring

00:18:36.360 and I was basically hooked. I was so hooked that I didn't really understand what I was signing up for

00:18:43.380 because the training for it after that was pretty difficult. I have to say seven years, but it was

00:18:49.260 worth it. I love every minute of it. And I'm still learning, of course, whereas at the very beginning

00:18:54.260 of a new inflection point in neurosurgery, which is understanding how the human brain works.

00:18:59.960 There was an epiphany you had at some point in residency, wasn't there?

00:19:03.280 Well, I think there was an epiphany about we have this access, really privileged access,

00:19:09.440 to use the information on what we call brain mapping, basically. We do the brain mapping because

00:19:15.520 we want to be very precise about how we're approaching, for example, a brain tumor or a

00:19:20.820 spot of the brain that's causing seizures. And we do the brain mapping so that we can map out the areas

00:19:25.400 that are really important for language or the ability to move your arm. That's what we call the

00:19:29.460 brain mapping. And we want to identify those areas so we can protect them during the surgery.

00:19:32.980 At the same time, do the maximal resection, like we spoke about earlier. The more that we can remove

00:19:39.840 the tumor, the longer their survival. The more that we can remove of the seizure zone, the more

00:19:45.100 likely someone is going to be cured of their epilepsy. But in neurosurgery, there can always be a cost.

00:19:52.240 And that cost would be paralysis or aphasia, which is a condition where you lose the ability to speak.

00:19:58.700 We're always trying to balance. Is it worth it to go that extra couple of millimeters versus not?

00:20:05.600 And in many cases, these are really profoundly important decisions that have to be made

00:20:10.040 right then and there in the operating room. And brain mapping is a way that we figure this out.

00:20:15.540 In the old days, Wilder Penfield, 150 years ago, 120 years ago, would use an electrical stimulator

00:20:22.100 and would apply it to the cortex. And that will temporarily activate or disrupt the function of

00:20:29.100 a specific part of the brain while someone is trying to speak or move. And that's traditionally

00:20:33.220 how we've done mapping. And that's one of the tools, the techniques we still use today

00:20:37.640 in order to make sure patients are safe during these procedures.

00:20:41.480 One of the revelations I had during residency, of course, is that I think we can do a lot more

00:20:46.900 than just applying stimulators to do the mapping. We've developed technologies to record from the

00:20:52.840 brain that allows us to not only do the mapping, but also is really the first window that we have

00:20:59.300 of understanding essentially how neurons work, how they convey information about words. For example,

00:21:06.440 like in the conversation we're having, there's a part of your brain in the temporal lobe,

00:21:10.340 which is right above your ear on both sides. The one on the left in particular,

00:21:13.900 you're right-handed. So 99% of right-handers are dominant for language on the left side.

00:21:21.300 And there's this one spot in the temporal lobe, which is just about two centimeters above your

00:21:25.600 left ear. That's processing all the words that I'm speaking to you right now. And so we've used

00:21:30.920 this technology not only to map those critical functions, that's where the science was, I would

00:21:37.380 say about 10 or 15 years ago, figuring out where these functions are in the brain. And now we've moved

00:21:42.900 the science to understanding how those areas work by a progressive evolution of new technologies to

00:21:50.520 get us to higher and higher resolution. And what I mean by that is we can measure the neural activity

00:21:56.600 of cells and cell populations and then link them to, for example, different consonants and bowels.

00:22:03.720 And that's been really an extremely exciting development in human neuroscience.

00:22:09.140 I want to back up and just have people understand how you even do awake surgery,

00:22:13.020 because the traditional way that surgery is done requires general anesthesia. And general anesthesia

00:22:20.740 typically requires three things. It requires one type of medication to blunt pain, another type of

00:22:28.000 medication to block memory, and another type of medication to paralyze you. Now, of course,

00:22:33.660 if things go wrong and things have gone wrong, sometimes patients are paralyzed, but they feel

00:22:38.760 pain, but they can't communicate it. And these are these catastrophic, but fortunately very rare events

00:22:42.940 that occur in anesthesia. But correct anesthesia is done where a patient has no sense of time or memory

00:22:48.780 of anything. They can't feel anything and they can't move, which means they're actually safe. That's to save

00:22:54.120 them as well. Help us understand how it is that you can do surgery without all three of those conditions

00:23:02.260 being present. Like I said in the beginning, this is extreme form of medicine, extreme form of surgery.

00:23:09.160 And the way that in a nutshell it can be done is the brain itself doesn't have any pain receptors.

00:23:16.180 So the pain receptors are the ones that are in our nerves, that are in the scalp, that are throughout

00:23:21.380 our body. These are actually the way that we perceive pain and touch. I think it's paradoxical

00:23:28.200 to many people at first, but the brain itself, which is processing that information in the body,

00:23:32.800 actually doesn't have those receptors itself. So the way that we typically do this is to numb

00:23:38.820 the scalp. We use things like when you go to the dentist's office, like lidocaine,

00:23:43.840 we can inject around the site of the incision. The bone, by the way, doesn't have any pain receptors

00:23:49.520 itself either. The membrane on top of the brain that we call the dura does have some pain receptors.

00:23:55.580 So sometimes we have to be sensitive around that and do some local anesthesia around the dura.

00:24:01.680 And interestingly, the brain tissue itself doesn't. There are some other areas like around the blood

00:24:07.100 vessels that can be sensitive. The membrane, the dura, is sensitive. So there are areas, but they can be

00:24:13.960 numb. And so this is a really important fact that allows us to do these surgeries awake when it's

00:24:20.800 necessary. So the patient is rolled, and I've seen these, and I just, I'm blanking on exactly the

00:24:25.360 procedure. So the patient is rolled into the OR. They're not intubated, correct?

00:24:30.960 Right. And they're never intubated during this.

00:24:32.540 Yeah. This is a patient who is laying there wide awake, no endotracheal tube.

00:24:38.540 Right.

00:24:39.180 Maybe a Foley catheter for some comfort?

00:24:41.080 Yes, Foley catheter. And usually Foley's are not that comfortable, but primarily so that we can

00:24:46.260 monitor the urine output during the surgery. You begin by making your incision, I mean,

00:24:51.860 drawing where you want to make your incision, and then literally just doing this as though it's a

00:24:55.280 local, like you're having a lipoma removed or something. You're literally just covering the

00:25:00.200 lidocaine and epinephrine across the scalp. You're bovying down to where you need to go.

00:25:05.900 So once you get to the bone, you can start to literally put a hole in and start to saw across

00:25:11.580 your holes. 0.84

00:25:12.420 Yeah. Just to add a little bit more detail to that, which is that usually the head is fixed. So it's

00:25:19.020 not like someone's sitting in a chair and we're just doing this while they're moving around, but we

00:25:23.440 have a head holder that fixes the head.

00:25:26.080 The patient is somewhat sedated and then you can lift the sedation.

00:25:28.740 Yeah.

00:25:29.240 Exactly right. And so we do a light level of sedation.

00:25:32.440 Like propofol or?

00:25:33.620 Like propofol, but at much, much lower dose. So it's not a general anesthesia dosing. It's a

00:25:39.060 very, very light dosing.

00:25:40.520 The party dose. And that way it allows the anesthesiologist to stop it when you say,

00:25:46.240 hey, because we might need an hour to get in. At that point, it's okay for them to be

00:25:50.200 in la-la land, but then I want it off once we have to get inside. 0.62

00:25:54.060 Absolutely. The period that someone is actually awake during the surgery is usually only an hour

00:25:59.020 or two, even if the surgery is like six or eight hours long. And primarily for comfort,

00:26:04.260 we'll do sedation at the very beginning. We'll have the sedation turned off so the patient can

00:26:08.620 be fully awake for the brain mapping.

00:26:11.780 And then you can ramp up the sedation to finish the procedure and close them. But again,

00:26:15.600 it's all done at the level of a colonoscopy, not the level of general anesthesia.

00:26:19.960 That's right.

00:26:20.120 Again, we were talking before the podcast, how my second month of general surgery, I did

00:26:25.000 neurosurgery, meaning we rotated through it. And one of the things that blew my mind was how it

00:26:32.360 sounds silly, but when you look at textbooks and you study the homunculus and you look at all of

00:26:37.060 the vasculature and then you actually look at the brain, it's just like any other organ. It's just

00:26:41.400 kind of a blah. It's like looking at the pancreas for the first time. You're like, that's it? How do I

00:26:46.700 know where everything is? I guess what makes it different is there's probably no part of the

00:26:51.720 body where the real estate matters so much. When you're operating on other parts of the body,

00:26:57.080 for example, if you're operating on the heart, you can really see what you're doing. You know

00:27:01.640 where the left anterior descending artery is. You know where the occlusion is. And even though it's

00:27:07.080 very technically complicated surgery, there is complete anatomic clarity of what is happening.

00:27:13.380 I think the thing that struck me the most the first time I saw neurosurgery was how the hell

00:27:20.500 do they know what they're doing? Like how many billion cells did we just lop off there? And

00:27:25.260 obviously this speaks to what you're talking about from a functional standpoint, but a lot of times

00:27:29.100 you're not doing that. So if a patient has a meningioma or some other tumor, how do you bracket

00:27:35.040 that trade-off? So do you sort of say, look, there are places where you never want to have a tumor.

00:27:39.640 For example, right above my left ear, that would be a really difficult place to have to resect because

00:27:45.340 the real estate is so precious. And that's where we're going to probably recommend doing an open

00:27:49.920 awake procedure to help guide us. Is that how you're using that?

00:27:54.160 That's exactly right. And so the real estate is critical. That's an understatement. But that being

00:28:00.000 said, there are some really expensive real estate in there. And there's also some cheaper real estate.

00:28:05.260 But that's, I guess that's my point, Eddie. It's like, that's sort of like telling me about

00:28:09.300 Manhattan. Like there's no cheap real estate in Manhattan. There happen to be areas that are

00:28:15.660 $10,000 a square foot, but there's probably nothing less than 4,000 a square foot.

00:28:20.160 So there's this popular idea that we only need 10% of our brain. I'm sure you've heard this.

00:28:25.360 I've heard this and I don't know what it means. It sounds like malarkey to me.

00:28:28.340 Right.

00:28:28.680 That might mean to stay alive, to respire. You might only need 10% of your brain.

00:28:32.580 So what it really means is that there is maybe about 10% or 15% that is very critical for our

00:28:40.500 basic functions, our ability to move, to talk, to see, et cetera. It's actually a lot more than that.

00:28:45.600 But it's also referring to this point that there are parts of our brain, actually, that are extremely

00:28:51.260 redundant with other parts of the brain. So the frontal lobes, for example. We do surgeries there

00:28:57.060 routinely, and oftentimes people really have no effect.

00:29:00.180 Even in terms of judgment, even in terms of-

00:29:03.040 Absolutely. Yeah.

00:29:04.240 Because we always think of the frontal lobe as where we have sort of executive function and where

00:29:08.580 we have the ability. We always joke, like one of my friends in med school, we said he had no frontal

00:29:12.780 lobe.

00:29:13.240 Exactly.

00:29:13.800 He just couldn't stop saying the most inappropriate things.

00:29:16.460 And don't get me wrong. The frontal lobe actually has a lot of critical function for executive

00:29:20.820 decision-making, impulse control, et cetera. But my point is that it's redundant, meaning that

00:29:26.660 different parts of the frontal lobe actually have similar purposes and similar role. And so for the

00:29:32.420 most part, a lot of our patients can accommodate a fairly large surgery, sometimes even removing the

00:29:37.740 entire frontal lobe.

00:29:38.580 Both sides?

00:29:39.720 No, not both sides. Really, usually these pathologies are only on one side.

00:29:43.460 If you took the entire frontal lobe from the left or right side, would there be a substantial

00:29:48.300 difference?

00:29:49.240 I've done that many times, just so you know. And it really depends on the case and scenario. If

00:29:55.100 someone's been having something that's slower growing there and there's been time for the brain

00:30:00.240 to reorganize, what we call plasticity, a lot of those functions will essentially no longer be

00:30:05.080 in that right frontal lobe and they've moved to the left side.

00:30:08.660 Wow. What is the mechanism by which that happens?

00:30:11.700 Time and function, meaning these things don't happen overnight. They take sometimes weeks and years.

00:30:18.880 But basically what happens is some neurons get lost over time and then others will compensate in

00:30:25.720 terms of that function.

00:30:26.940 But how does that actually happen? So let's posit that we have a slow growing lesion in the left

00:30:32.960 frontal lobe. What is the left frontal lobe doing to communicate with the right frontal lobe

00:30:38.320 to say, hey, these neurons are being compromised. Their function is deteriorating. You guys need to

00:30:44.920 pick up the slack. How is that message being transmitted?

00:30:47.880 Part of it is that both parts of the frontal lobe for people, most people are both doing the

00:30:53.660 function most of the time. So it's not like it's just transferring the information. It's that both

00:30:59.940 sides were originally involved in those functions. And then one side gets weaker and the other one has

00:31:06.580 to pick up that slack. At a cellular level, this is what we call synaptic plasticity. The weights,

00:31:12.680 you know, essentially make up who we are. These are just the weights that neurons use to communicate

00:31:18.600 with one another. All of our learning is towards shaping that weighting of synapses that occur

00:31:25.320 where neurons touch each other. And that can happen. That can change throughout our life.

00:31:30.860 Every time we learn a new word, those are new synapses that have formed that were never there,

00:31:35.840 new connections. Precisely from the left to the right side, there is this structure that we call

00:31:40.720 the corpus callosum. It's an information highway that connects the left part of our brain from the

00:31:46.840 right side. There was a Nobel Prize, Roger Sperry, who'd done really incredible early experiments

00:31:51.600 describing patients who had surgeries where you split that. And in certain instances, you have this

00:31:57.800 phenomena where people essentially like have two functioning brains, but they're not communicating to

00:32:03.380 each other. So it does require that there is this connection between the two areas where they're being

00:32:08.500 reorganized. Now, outside of epilepsy, why else would the corpus be severed?

00:32:13.620 That's really the main one that we use it for. Can you describe how patients that undergo that

00:32:19.620 procedure behave? It's very fascinating. So there's a phenomenon that we call a dissociation

00:32:26.060 syndrome. The clinical indication, the medical indication for why someone would undergo this

00:32:32.140 nowadays is that some patients with seizures have severe seizures where they fall down,

00:32:39.000 what we call drop attacks. And usually what that means is that the seizure is spreading so quickly

00:32:44.260 across the brain that people lose the tone in the body and then basically fall. And why that becomes a

00:32:51.980 problem, the injury of the seizures, is that people actually injure themselves. So it's not uncommon for

00:32:57.080 people who have these kinds of seizures actually to be wearing helmets all the time because they're at such

00:33:02.140 risk of falling. These are just medically recalcitrant seizures. They cannot be prevented

00:33:07.920 with any degree of... That's right. That's absolutely right.

00:33:10.960 I see. Without any kind of medication. And these are particularly ones that there's not one small

00:33:15.980 spot that's causing the seizure. It's the whole half or quadrant of the brain. But the problem is

00:33:21.580 what a seizure is is basically when you have this very uncontrolled synchrony of a large mass of the

00:33:28.940 brain cells. So normally, if you think about the brain and the neurons within the brain, it's like

00:33:35.500 people in a stadium. They're having their individual conversations. That's the way the brain normally

00:33:40.080 works. But let's say all of a sudden everyone's doing a wave and something hyper-coordinated.

00:33:45.720 All of those normal conversations are now gone. The brain has now become hijacked by this other

00:33:53.100 phenomenon where everything has become very coordinated. That's why people lose consciousness

00:33:57.640 because all the normal function is basically shut down. And the way that it can become synchronized is

00:34:05.240 through its connectivity. Every cell connecting to its adjacent cell and every cell connected to all the

00:34:12.000 other cells in the brain through the things like the corpus callosum that connects left to the right.

00:34:17.700 And so when you get that hyper-synchrony, people can essentially lose consciousness almost

00:34:21.840 instantaneously. So one of the reasons historically why the corpus callosum was invented in the very

00:34:28.940 first place was to sever the connection between the left and right hemisphere.

00:34:33.620 Which doesn't stop the seizure. It just limits the spread.

00:34:36.520 That's exactly right. It doesn't stop the seizure. But what it does is stops the propagation,

00:34:41.360 the very fast propagation of the seizure from one side to the other. And in order for someone to

00:34:47.800 lose consciousness, you basically have to have both sides of the hemisphere or a deeper structure like

00:34:52.880 the thalamus. That's basically, in order to have consciousness, you have to have both hemispheres out

00:34:59.660 in order to lose consciousness or a deeper structure in the thalamus or something like that.

00:35:04.580 And so these kind of drop attack seizures are ones that people black out, fall. And one way you can

00:35:12.460 actually dramatically stop a lot of that is disconnecting the left from the right hemisphere.

00:35:17.660 And so how does that patient, how is their life different aside from the fact that hopefully their

00:35:21.680 drop seizures are gone? What's the change in the way that their left and right behave now disconnected?

00:35:28.780 Most of the time when we do this nowadays, we disconnect about the anterior two-thirds of the

00:35:33.880 corpus callosum. And the reason why we don't typically transect the whole corpus callosum

00:35:39.540 is because of some of the side effects that people can have. And it's what we call a dissociation

00:35:44.360 syndrome, where you can basically have a dissociation between what the left brain is doing

00:35:50.120 in the right brain. So for example, someone essentially feeling something on the right hand,

00:35:56.100 which is processed by the left part of the brain, and the right part of the brain really having no

00:36:01.020 awareness of what's going on. People can get by with that, but it does affect how they can get along.

00:36:07.360 So nowadays, we try to just do the front part of it and leave the back part that helps reduce some

00:36:14.040 of those side effects. But does it solve the initial problem? Of the seizures? Yes. No. It

00:36:19.320 doesn't cure the seizures, but it really just stops the propagation of the seizures so that people

00:36:23.700 don't lose consciousness. Sorry, by leaving one-third in the posterior still adjacent, it prevents the

00:36:28.580 propagation across the hemispheres? In many, many cases. In some cases, nowadays still,

00:36:34.280 we have to do a total callusotomy. It's a very delicate surgery because the corpus callosum that

00:36:40.160 connects the left and the right, it's not sitting on the top of the brain. It's actually- It's very

00:36:43.940 deep. It's very deep. So to get there, you actually have to physically separate the left and right

00:36:50.480 hemisphere. We do that from the top. We do a craniotomy that's centered over the midline, 0.85

00:36:55.140 but it can't be right over the midline because we have a large draining vein there called the

00:36:58.780 superior sagittal sinus. So we have to either choose the left side or the right side. And then we have to

00:37:04.280 carefully separate the left from the right and through that narrow corridor, transect those

00:37:09.920 connections. And what's directly underneath it that prevents you from just running that bovie a

00:37:16.160 little too hot? Oh, well, there is a blood vessel that runs along the top of the corpus callosum.

00:37:21.980 And that's actually the most critical part of the surgery is that we separate those two branches,

00:37:27.400 those pericolosal arteries. Those arteries are really important because they supply the part of

00:37:32.280 the brain, the medial frontal cortex. And the part of that that is part of the motor cortex

00:37:37.740 actually is what supplies and controls our legs. So if you have a stroke, let's say, as a side effect

00:37:44.320 or complication of that procedure, then someone would be paralyzed in the leg. So it's a fun,

00:37:50.480 delicate surgery. It's amazing to see that exposure of the corpus callosum. It's glistening white.

00:37:57.360 That's how we can see it. And it's very distinct from the cortex because it's really clear white.

00:38:02.200 That white comes from the myelin. It's a heavily, heavily myelinated structure because it's conveying

00:38:09.020 information from the left side of our brain to the right side on the order of milliseconds,

00:38:13.400 a super fast connection.

00:38:15.740 Do you ever think philosophically about what the implications are for human consciousness

00:38:19.540 by the fact that you can do a complete transection of the corpus and seemingly produce two people?

00:38:26.280 That's right. Or potentially two consciousnesses.

00:38:30.080 Yeah. What does that mean?

00:38:31.520 Well, I think that goes back to a harder question of how do you define what consciousness is in the

00:38:37.520 first place? And this is where there's a lot of philosophical debate about that.

00:38:41.060 Do you trouble yourself with such debates? I mean,

00:38:42.920 Not really.

00:38:43.480 I can't because I can't wrap my head around it.

00:38:46.060 Yeah.

00:38:46.280 It's above my pay grade.

00:38:47.740 I do think about it from the practical clinical perspective,

00:38:51.420 not so much the philosophical. The clinical one is like, is a patient in a coma or not? And why?

00:38:58.420 And how do we get them out of that? Traumatic brain injury, certain strokes, epilepsy, etc.

00:39:04.760 We think about consciousness from that perspective, literally all the time, every day.

00:39:08.880 But from the philosophical, I don't lose a lot of sleep over that one.

00:39:13.460 I lose a little bit of sleep over it. Let's talk a little bit about a brain computer interface.

00:39:17.780 You've mentioned it already. I hope I'm not insulting people when I say this,

00:39:22.320 but if we're going to be brutally honest, we should at least acknowledge that medicine to date

00:39:27.320 has been pretty unimpressive when it comes to treating neurodegenerative disease. So whether we're

00:39:32.560 talking about Alzheimer's disease or even other non-degenerative forms of dementia, whether we're

00:39:39.080 talking about Parkinson's disease, Lou Gehrig's disease, I mean, we just don't seem to be able

00:39:44.040 to treat these diseases. So whatever medications we throw at these things, maybe in the case of

00:39:49.160 Parkinson's disease, we can delay progression a little bit. But would you agree with that

00:39:53.820 assessment that the traditional approach to treating these diseases has been largely unsuccessful?

00:39:58.900 I would largely agree with a caveat that I think a lot of progress is being made to understand

00:40:04.900 what's going on. And from that, I think there's a lot of promising therapies.

00:40:08.280 I would generally agree that within neurology and neurosurgery, traditionally, therapy has really

00:40:16.000 been designed to stop things from getting worse or slowing progression. Replacing function has never

00:40:23.780 really been possible until very recently.

00:40:26.280 Yeah. I guess what I want to understand from you, because I think this is something that you know more

00:40:32.420 about than anyone, or certainly among the people who would know the most about it, is do we need to

00:40:37.780 revisit our approach to these diseases more from an engineering perspective than from a peripherally

00:40:43.700 administered medication perspective? So the traditional approach to treating disease would be

00:40:50.360 medication. You take an injection, you take a pill, you take something, and you hope that enough of it

00:40:57.300 gets across the blood-brain barrier and it starts to treat the condition at hand. But it's hard to look

00:41:02.800 at a patient with Parkinson's disease, which is a motor defect disease, admittedly that stems from the

00:41:09.440 CNS, and not at least think this is a functional condition. Why isn't there, or is there, an engineering

00:41:16.980 approach that could be taken to this?

00:41:19.340 It's a good question. And let me put it this way. Medications are always going to be a really

00:41:25.860 important goal, not only to reduce symptoms, but hopefully find cures. But there's this whole

00:41:33.120 other class of therapies that are coming online that have to do with this other property of brain

00:41:41.100 cells, which is very different than, let's say, the pancreas or the liver. It's the electrical side of

00:41:47.520 the equation. So there's the chemical and biological side, but then there's this electrical side. And the

00:41:53.760 brain is an electrical organ. Our thoughts are really dependent on these electrochemical kind of

00:42:01.380 processes that happen at individual neurons and the collection of them. So there's this large and

00:42:08.880 growing field that we call neural engineering that is really trying to use computers, sensors, chips,

00:42:17.260 in order to interpret eavesdrop on how the neurons are signaling to each other, regardless of their

00:42:25.920 pathology or the biology. But just what are they saying to one another? Can we eavesdrop on that?

00:42:31.800 Can we interpret it? Can we decode it? And then more importantly, can we use that information actually

00:42:37.500 to guide more normal signaling? Why this is potentially important is that at the end of the day,

00:42:43.140 the function is from that electrical activity. It's the propagation of those action potential by

00:42:49.340 neurons, which gives rise to our thoughts, our ability to communicate, our ability to walk,

00:42:53.740 move our arm. Without that, it's not there. So I would say neuroengineering as a complement

00:42:59.980 to the biological or pharmaceutical approaches.

00:43:04.180 Eddie, if you and I were mapped simultaneously together, this question might not even make sense.

00:43:10.120 So please feel free to adjust the question to make it logical. But I think you'll understand

00:43:14.160 what I'm trying to ask. All of the electrical activity of my brain could be mapped to a computer

00:43:20.720 and the same could be done with yours. And we were thinking the same thing. So it was an experiment where

00:43:26.260 we were both told to think the same thing. Peter, Eddie, we both want you to think about sitting on a

00:43:31.600 beach with your feet in the sand. It's hot. As descriptive as you want it to be, would the outputs on the

00:43:37.580 computer screen be similar? Would the computer be able to appreciate similar electrical output?

00:43:44.860 Or could two people that are doing the best job they can to have the same thoughts not be able to

00:43:50.240 produce that? In other words, is there a one-to-one map of thought to electrical activity?

00:43:54.440 The short answer is actually both. So in that example that you gave, if both of us are looking at

00:44:01.520 the same picture of a scene on the beach, yes, the same part of our brain is going to be processing

00:44:07.580 those images in the very back of the brain that we call the occipital lobe. The primary visual cortex

00:44:13.060 is going to be parsing that space into what we call retinotopic space, like essentially where those

00:44:19.700 different pixels are located in the image. That part is going to be highly conserved, not identical,

00:44:25.460 but highly conserved between your brain and mine. It's where these computations go further upstream,

00:44:33.880 where they become much, much more differentiated, much more specific to our brain, much more dependent

00:44:41.720 actually on our history, history of thoughts, our personality, everything that's interacting with

00:44:48.440 the rest of the brain. I'll give you a great example. The way that you may hear Spanish or French

00:44:54.940 or German is going to be very different than someone who is a native speaker of those languages.

00:45:01.360 Your brain is going to process some of those sounds. You'll hear them, but you're not going

00:45:05.640 to be able to pick out the words very easily. Or the way I hear an engine versus the way my wife

00:45:10.700 hears an engine. I love the sound. She's mildly annoyed by the sound. We're hearing the same thing.

00:45:16.520 Absolutely. Yeah. So there are parts that are going to be very similar, like how we process

00:45:21.420 some of the sensory attributes. And then the further you go deep into the system, the more it

00:45:26.900 becomes very, very tailored. Some of this is hardwired. The way that our visual system early

00:45:33.780 on, a lot of it is hardwiring. It's heavily influenced by what we see.

00:45:38.980 Seeing a snake should automatically produce a negative response that doesn't have to be learned

00:45:44.620 in theory, I assume. Evolution has probably hardwired us for that.

00:45:47.740 There are some things that are instinctual. Certain odors, we'd be hardwired not to like

00:45:52.120 things that are going to smell rotten or something like that.

00:45:54.040 Oh yeah, absolutely. There's a lot of those things that are very intuitive.

00:45:57.800 Now going back to the, we're showing you the picture of the beach, how much do you change over

00:46:02.700 time? So if we did that experiment when you were 10, 20, 30, 40, 50, would that also change?

00:46:10.000 It will change, but less. I think over time, these things become refined and over time actually

00:46:16.100 lose their refinement. So as we age, some of those representations actually become less clear.

00:46:21.580 If we talk about hearing, for example, there are a lot of people in the population over time,

00:46:27.780 it's very hard for them to be in a crowded restaurant where there's a lot of background

00:46:31.740 noise, competing conversations going on. And yet these individuals can have perfect hearing.

00:46:38.080 So signal processing is becoming the problem.

00:46:39.960 That's exactly right. It's not an ear problem per se. It's a perceptual problem and largely in the

00:46:46.440 brain. And that has to do with how that information, like the fidelity of those signals.

00:46:51.240 Is that a harbinger of something bad?

00:46:53.880 Independently, no. But we do know that when people have that problem, they tend to be more

00:46:59.260 socially isolated. So there's a lot of secondary things that happen. We do know that when people

00:47:05.040 have hearing loss and it's unrecognized, a lot of people have unrecognized hearing loss. And what

00:47:11.920 people don't fully appreciate is that if you don't have access to communication, to conversation,

00:47:20.320 your brain is not getting those same signals. It's becoming deprived. And what we do know through

00:47:26.640 many studies now is that the cognitive effects of that hearing loss actually can be quite profound.

00:47:32.800 It accelerates age-related memory loss.

00:47:36.320 We actually internally in our practice believe that that is causal to cognitive decline. Now,

00:47:42.560 there was a study that came out about two years ago that suggested it wasn't. Although the study

00:47:47.920 had a partial retraction, the methodology was a little flawed. But to my knowledge, I don't know

00:47:52.720 if it's been repeated. The question being, of course, is if you correct hearing loss, let's say you

00:47:58.720 randomize a group of people with early MCI and you correct hearing loss, do you correct or prevent or

00:48:04.960 reverse it? And again, if there's causality there, you would expect that you would. What about with

00:48:10.480 visual? As people get older and they develop cataracts or things of those nature and their

00:48:15.440 visual acuity goes down, does it have the same effect on depriving them of enough neural stimulation

00:48:22.640 to maintainer? Is it not as much because it's not a language issue?

00:48:25.200 It's not as much. And I'm less aware in the prevalence of something that's age-related,

00:48:30.960 just in the visual cortex, for example.

00:48:32.800 So let's go back to brain-computer interface. How would you explain this to somebody at a party

00:48:37.440 if they said, that sounds pretty high-tech, but what is it?

00:48:40.080 Okay. Let's just break apart the terms. Brain refers to really any kind of thing that interfaces

00:48:47.440 with the cortex or the deeper structures. The computer is a digital device on the outside.

00:48:54.640 A lot of people now call this BCI, brain-computer interface for short. It's a very messy term

00:49:01.040 because it could mean a lot of different things. I think in a nutshell, what it means is for most

00:49:06.880 people, a system that is recording from the brain, whether it's non-invasive from the scalp

00:49:13.840 or something that's fully invasive within the brain itself, and connecting those signals to a computer

00:49:20.560 that analyzes the signal and then does something with it. In many cases of BCI research, the application

00:49:28.080 is, for example, to remove a computer cursor, or the research that we've done is to replace speech

00:49:35.680 words for someone who's severely paralyzed and unable to talk anymore. It's about interpreting

00:49:42.960 brain signals and then using a computer to interpret those signals and then transform

00:49:47.920 them into a form that's useful to us. So in that example you gave, you're describing a patient with

00:49:54.080 aphasia who can't speak? Let me be very specific about that. A lot of the work that we've done

00:50:00.400 is on people that have a severe form of paralysis. And aphasia we typically refer to as someone who's

00:50:05.760 got, let's say, a stroke in the language centers of the brain. Where we focus recently is on patients

00:50:12.160 that have a severe form of paralysis like ALS. So there the problem is they have largely normal language,

00:50:19.200 but they can't get those... Can't get the motor signal out.

00:50:21.680 Can't get the motor signal out to the vocal tract, the lips, the tongue, the jaw, the larynx. Those

00:50:28.160 descending fibers are severely affected by ALS. They degenerate. And that's why people progressively

00:50:34.560 become paralyzed and lose ability to speak. An important part of that is that they lose ability

00:50:39.200 to speak, but they still have full cognition. Yeah. And for that individual, by attaching a

00:50:45.440 computer to their brain, you're able to hopefully extract in written cursive text, whatever across

00:50:53.760 the computer screen, what they're wishing to say. That's right.

00:50:57.920 Okay. Let's talk about how that could possibly be done. You mentioned earlier, there are at least

00:51:04.160 two broad ways to extract that information. A non-invasive way where presumably you're putting

00:51:08.960 electrodes all over a head that's as well shorn as mine. Alternatively, a very invasive way where

00:51:16.000 you actually remove the scalp and you lay these things on the cortex itself, correct?

00:51:21.520 Yeah. So the range would be EEG, which is where sensors are placed on the scalp directly,

00:51:29.520 recording non-invasively. You can remove them at any time. And then the far other extreme is

00:51:36.800 electrodes that are actually placed into the brain, the most invasive. ECOG.

00:51:41.040 ECOG would be electrodes that are on the brain surface. That's short for electrocorticography.

00:51:45.600 And that's where we've done the vast majority of our work in my lab.

00:51:48.880 And that's just placed on the surface of the cortex, under the dura on the cortex.

00:51:53.120 That's absolutely correct. So what's nice about that is that you don't have the injury to the brain

00:51:59.520 itself from the insertion of the electrodes. It's a stable recording over time. We now use

00:52:05.600 ECOG devices to essentially help people with seizures, for example, where you can basically

00:52:11.440 have a pacemaker now that records from the brain surface and then stimulates to help stop the seizures.

00:52:16.320 And this is a fully implanted device?

00:52:18.640 It's moving towards that. So the work that we've done in our clinical trial

00:52:23.360 is using an array that's surgically placed, but connected through a port. We call it a

00:52:29.040 percutaneous port because it's actually physically attached and anchored to the skull. An array on

00:52:35.520 the brain, it comes out the dura and is anchored in the skull.

00:52:39.680 Where does the port exit the body?

00:52:41.520 Right on the top of the scalp.

00:52:43.040 How do you prevent infection with such a close...

00:52:45.600 Yeah, it's a really good question. And that's what the main problem with

00:52:49.040 this early prototype BCI. And it's our group, there's other groups around the world that are

00:52:54.160 using similar things primarily to show if it's possible actually to code brain activity for useful

00:53:02.560 purposes. So what's happening right now in the field is a lot of these technologies are now going

00:53:09.280 to become wireless over time. But you're absolutely right. One of the main reasons

00:53:13.520 is that we want to move away from the percutaneous, we want to move away from the ports, which are

00:53:18.240 infection risks on top of other problems, and move to things that are fully implantable, fully wireless.

00:53:24.960 How long do you feel you're away from that?

00:53:26.800 Yeah. Basically about a year.

00:53:28.320 Oh, okay.

00:53:29.360 We've been working on it for quite some time. And so it's a really interesting time where we're

00:53:33.840 seeing a convergence of what's possible with electrical engineering, high bandwidth, wireless,

00:53:40.560 perhaps way beyond what we can do with Bluetooth, advanced electronics that now allows to print some

00:53:45.840 of these sensors on a substrate that is thinner than a piece of paper, really, really small, and on a substrate

00:53:54.080 that can conform to the convolutions, the different peaks and valleys of the human cortical surface.

00:54:00.640 So what is the trade-off between ECOG and sensors inserted directly in the brain? What's the resolution difference?

00:54:08.640 Well, that's a very important question that we and many other people are trying to figure out right now.

00:54:14.800 Most of the time when people are putting an electrode into the brain, there has to be some gain for that.

00:54:21.120 And usually that's for recording a higher resolution, usually trying to record the activity of

00:54:28.400 single neurons, a single cell.

00:54:30.880 How can you isolate a single cell?

00:54:33.600 Really small electrodes, super small electrodes. We do a lot of work with this in our research.

00:54:39.440 The challenge for the field has just been that it's very hard to stably record from single cells

00:54:46.720 more than a couple of hours or days.

00:54:49.680 So that's one challenge.

00:54:50.720 The other challenge is that when you put the electrodes into the brain,

00:54:53.920 it can create a reaction.

00:54:56.240 Those glial cells that we talked about at the very beginning, those support cells,

00:54:59.600 they actually have immune function as well. They detect that there is a foreign body and they'll

00:55:04.960 activate and they'll react to it, create a scar around the electrodes.

00:55:10.080 So the advantage of having electrodes in the brain to do this very microscopic kind of recording is that

00:55:16.800 you can get a finer signal from single cells. The disadvantage is that it can create a reaction

00:55:22.720 that reduces the fidelity of those signals over time. One of the reasons we've been most interested in

00:55:28.720 using these sensors on the brain cortex is that we've learned over time that if you don't have

00:55:33.440 the electrodes penetrating through the peel surface of the cortex, that's the outermost

00:55:38.240 very thin membrane that's covering the cortex. If you don't have anything going through that,

00:55:42.720 you can avoid a lot of those immune reactions, avoid a lot of that scarring, preserve the function

00:55:48.080 that's underlying. But this is something that we're actively trying to understand.

00:55:52.240 So just to give me a sense of magnitude, if EEG on the surface is one, one unit of resolution,

00:56:01.280 what would ECOG be? And what would implanted electrode be? Is it one, 10, 100? What's the scale

00:56:08.160 at which you're thinking about resolution? Well, I would say from the scalp, let's say we just

00:56:13.120 arbitrarily call that one. And then you think about what you could do with this ECOG.

00:56:17.920 ECOG. I think we're really talking about, let's say, a thousand times better resolution. We've

00:56:25.120 been able to answer very fundamental questions actually about how the brain works using those

00:56:30.240 kind of surface recordings in a way that's impossible with surface scalp electrodes. And

00:56:35.840 then once you go take that further to single neurons and you've got another resolution, probably

00:56:39.920 takes to 5,000. The big jump is just going from an EEG to an ECOG.

00:56:45.840 Right. Directly to the brain.

00:56:46.880 That's a three log change.

00:56:48.480 Right. Whereas you're a 5X change going deeper.

00:56:51.120 Right. And so one of the reasons for that is the skull and the scalp are a major loss of signal.

00:56:58.080 The signals are small to begin with. So once you're trying to interpret them through the skull or

00:57:04.000 scalp, they're basically gone and very diffuse too. So trying to understand like where they came from

00:57:09.440 in any precise way is almost impossible. When you're recording directly on the surface,

00:57:14.240 you're basically at the source itself. The cellular level, the single cell recordings are terrific

00:57:20.960 for trying to understand that ultra fine resolution, primarily in the case that we use them for,

00:57:25.600 is for research, primarily to understand what's happening at those units. But still to this day,

00:57:31.200 there's really no way that you can chronically and stably record from the same cells.

00:57:36.800 Because of the immune reaction?

00:57:38.800 Also because of how fine of a problem it is, how precise it has to be. We're talking about a single

00:57:46.480 cell, a couple of microns in diameter, and you've got an electrode. Any micromotion at all,

00:57:53.680 anything changes that. And so typically what we see with a lot of those systems that record from those,

00:57:59.120 is there's a lot of turnover from day to day or hour to hour.

00:58:02.320 Meaning you're drifting between which neuron you're recording in?

00:58:05.440 That's exactly. Yes.

00:58:06.880 Does that imply then that, I mean, just taking a step back, you said 82 billion neurons in our brain.

00:58:13.680 So you put the probe into one and it moves over to the next one and the next one and the next one.

00:58:20.640 One like me would naively assume they're all the same. Those are like three row homes that are

00:58:26.720 basically all identical on the upper west side. We're not talking about Tribeca here. Does it

00:58:32.880 really matter if the probe moves between those three? It sounds like the answer is yes, but I'm

00:58:37.440 curious as to why. The answer is yes. Because we now know that cells that are right next to each

00:58:43.200 other can have some very different information. Now, that being said, when you go through a column

00:58:50.320 of the cortex, the column is this vertical organization. Typically what we're thinking

00:58:54.960 about when we look at is the two dimension of the surface. But there's a third dimension of

00:58:59.840 information processing, which is the different layers of the cortex, we call the lamina. And

00:59:05.200 typically in some of the sensory areas, for example, if you put an electrode, it's primarily going to be

00:59:10.880 tuned to the same information across those different neurons across that depth. So you're right,

00:59:16.880 in certain areas, you may have neurons that are tuned to the exact identical thing. And for decoding

00:59:22.960 purposes, that may actually not be a big deal to have it not very stationary over time. In other

00:59:29.120 instances where you're trying to code from areas where it's a lot more intermix, it could have really

00:59:33.680 profound implications where you have to recalibrate the algorithms that the machine is doing to

00:59:39.440 interpret the signals every couple of hours or days. So when you do ECOG, how do you direct the

00:59:47.600 sensor at the part of the brain you want to go to? Because I assume it has to be far more nuanced than

00:59:53.040 just where you slap it on the cortex. Yeah, we're definitely getting into details. And I love that,

00:59:58.160 Peter, about you. You're not afraid of getting into the details. This alludes to one of the things we were

01:00:03.600 talking about earlier. The part of my brain and the part of your brain that is responsible for

01:00:09.040 speaking, especially in the motor control part, is largely the same. It's in the same ballpark.

01:00:14.800 There's a lot of variation when we come down to the details, the microgeography. But it's in the same,

01:00:21.840 largely in the same city, if we're talking about geography. Where your house is versus my house,

01:00:27.840 etc., that's going to be a little bit varied within that. One of the nice features about using ECOG or

01:00:33.920 electrocorticography is that you can put an array over that entire area safely, and you can sample

01:00:39.920 very, very densely across the entire city, let's say. And so it doesn't really matter, actually,

01:00:45.440 at the end of the day, if one person's there and the others, basically you're going to cover it.

01:00:49.280 That seems to actually be a feature, not a bug, right?

01:00:52.160 Absolutely.

01:00:52.480 And the bug is you give up the resolution at what's happening in the kitchen of that house,

01:00:58.160 but you now get to look at all the houses. Exactly. And you get to do it in a way that's

01:01:03.440 very safe and scalable. I mean, the biggest thing you give up is 80% of the resolution, roughly.

01:01:08.240 Right.

01:01:08.720 Yeah. So with ECOG, tell me how many words per minute you could capture from a patient with ALS.

01:01:19.760 What we did in our clinical trial at UCSF.

01:01:23.520 Was this the 2023 paper?

01:01:25.280 Yes.

01:01:25.520 Okay. This is the nature paper.

01:01:27.120 That's right. We published a paper in 2023. We worked with a participant named Ann. She had a very

01:01:34.880 severe brain stem stroke about 20 years ago.

01:01:39.040 How old was Ann?

01:01:40.560 She was in her 20s. It wasn't long after she had gotten married. Just a couple months after

01:01:46.080 her second daughter was born. She was playing volleyball with her friends, collapsed, taken to

01:01:52.640 the hospital. She survived the injury.

01:01:55.840 Was it a vertebral artery dissection?

01:01:58.240 Yes, that's exactly right.

01:01:59.680 Yeah.

01:01:59.920 Okay. This is just such a scary thing.

01:02:01.920 Your memory from medical school is actually pretty good. This is impressive. But back to Ann,

01:02:07.200 absolutely devastating. Just so people understand, vertebral artery, everyone's heard of the

01:02:12.480 carotid arteries. Okay. The carotid arteries come up through our neck and they primarily give the

01:02:17.600 blood flow to the front part of the brain. The vertebral artery is an equally, if not more important,

01:02:24.240 set of arteries that supply the brain stem, which connects the brain to the spinal cord,

01:02:30.000 and the back of the brain. So we have these two pairs of really important blood vessels that come to

01:02:36.000 our brain. The carotid and then the vertebral arteries. And Ann had an injury while she was playing.

01:02:42.640 And it was really just unfortunate, but she had this stroke in the vertebral artery that blocked the

01:02:48.320 blood supply to the brain stem. So functionally, what this means...

01:02:52.800 Can I ask a naive question? Does it have to be bilateral to cause the injury? Or if it happens on

01:02:58.720 one side, can the other side not perfuse around the circle of Willis? Why does that injury happen?

01:03:03.840 Just to be even more precise about this, unlike the carotid artery, the vertebral arteries,

01:03:09.040 you have a left and right vertebral artery. They come up through your neck and then they go through

01:03:14.560 the base of the skull, through the foramen magnum, essentially where the spinal cord

01:03:18.720 is coming through the base of the skull. When they enter the skull, they become one artery. It's called

01:03:25.680 the basilar artery. And the basilar artery and the small perforating arteries that come off that supply

01:03:32.320 the brain stem are absolutely critical. So it depends where the dissection occurs.

01:03:37.200 Exactly. If it occurs before the bifurcation,

01:03:39.440 you're probably fine. If it occurs above the bifurcation or where they join, it's not the

01:03:44.160 bifurcation, but yes. Yeah, yeah, yeah.

01:03:46.160 Actually, you're right. There are many cases. In fact, sometimes we, for various reasons,

01:03:51.200 actually have to occlude a vertebral artery. And then the other one, just collateral,

01:03:56.320 gives the collateral flow. The basilar, however, doesn't have that kind of...

01:04:00.160 Yeah, it's mission critical. It doesn't have that insurance policy, no backup. It's such a critical

01:04:04.880 structure. And when there's a problem there, it's usually actually like terminal.

01:04:10.640 Ann survived this stroke. She was left quadriplegic, meaning she couldn't move her arms and legs.

01:04:18.560 But in addition to that, she couldn't speak because the nerves that come through the brain,

01:04:24.720 down through the brain stem and go to the cranial nerves, which supply the vocal tract,

01:04:29.040 those were also directly affected. And yet, just again, you'll have to pardon my profound

01:04:35.520 ignorance. Those would be lower cranial nerves. That's right.

01:04:39.600 Three, four, five, the ones for the diaphragm were intact, so she could still breathe on her own.

01:04:45.680 But what is it? Seven, eight, nine would have been compromised, which is why she couldn't speak

01:04:50.640 or something in that neighborhood. Yeah.

01:04:52.320 So it's the cranial nerves in particular around the lower ones and those distributions that allow

01:05:00.240 the control of the tongue. That's the hypoglossal nerve. That's number 12.

01:05:04.160 It's number 10, the vagus. But it's not precisely the nerves. It's actually the brain stem nuclei.

01:05:09.680 Where the nerves originate. That's exactly right.

01:05:11.680 And that's not something one predicts from any type of stroke. It's simply the nature of what part

01:05:16.720 of the brain stem was affected. That's exactly right.

01:05:18.560 And was the paralysis a result of her cerebellum also having infarcts?

01:05:23.520 No. No, it was all brain stem related. Just brain stem.

01:05:26.240 Good God. Yeah, precisely the part that we call the ponds.

01:05:29.680 Devastating. Devastating. So it's for 20 years,

01:05:33.520 Anne is now in her 40s in a wheelchair, unable to speak.

01:05:37.280 Right. So I think some important things about this are that it was actually about 18 years after

01:05:44.960 her stroke that she decided to participate in our trial. And we had talked actually a year earlier.

01:05:51.760 She said, I really want to wait to participate in this trial because I want to wait until my daughter

01:05:56.560 graduates and then I can do this with you guys. And I assume she said that because the risk of

01:06:02.080 something catastrophic happening was high enough that she felt she needed to wait.

01:06:05.920 Well, she's a mom. She wanted to be there for a daughter and she had a year before the 0.99

01:06:11.200 graduation. And she reached out to us because we had an earlier participant also with a brain stem

01:06:16.800 stroke that we treated, that we did this trial that I'm about to describe. And she read about it.

01:06:21.840 And so she reached out to us. How did she communicate at that point in time?

01:06:25.200 So the main way that she communicates is through devices that can track her eye movements. Those are

01:06:31.280 translated to a pointer that can point on a screen to individual letters or words. And so it's a very

01:06:37.280 painstaking way of communicating. One thing I've learned about Anne is she's just a tremendous

01:06:43.440 person, positive person. She's just a force of nature. She recently actually used that same

01:06:51.280 system to write a book chapter. Just incredible. Okay, so we started this trial called the BRAVO trial.

01:06:58.480 It's something that we worked with the FDA very closely to get approved because it requires a brain

01:07:04.880 surgery. It requires this percutaneous port that we talked about. The reason we were able to get it

01:07:10.640 approved in this form was that a lot of the components that we were using were actually

01:07:16.960 existing medical materials. So the safety of it was largely already known in terms of its

01:07:23.120 biocompatibility, its biostability. What was not known is that if someone has not spoken for a decade

01:07:31.360 or two, whether or not those parts of the brain actually would still work.

01:07:35.120 Yeah, it's really interesting because we know that if a person loses their sight

01:07:43.040 for X number of years, I'm guessing that the occipital lobe doesn't work the same way. It's not

01:07:50.480 processing the information. So I don't know that it actually physically atrophies, but I'm guessing

01:07:55.600 that the neurons aren't firing the same way, right? That's right.

01:07:58.800 So what would be interesting is we just don't know if Anne's inner monologue is still happening

01:08:04.960 the same way.

01:08:05.840 Right.

01:08:06.320 That's a very interesting question.

01:08:08.480 And I think that ultimately that was the biggest risk, actually. There's a lot of emphasis on the

01:08:16.080 technology, but the basic biology of how the brain works and whether that information is still being

01:08:23.040 processed, I think are really the more important ones, actually. And so what we did was we did a

01:08:29.360 surgery where we implanted an array of 253 ECOG sensors. These are the sensors that are densely spaced.

01:08:37.680 How many?

01:08:38.160 253. So we're talking about something about the surface area of a credit card, and it's

01:08:45.760 filled with electrode sensors that are spaced about three millimeters apart. Each sensor is about a

01:08:52.160 millimeter in diameter. And so basically you've got this credit card sized array that was placed on

01:08:58.640 the part of her brain that processes words, in particular the motor production of the words,

01:09:04.960 the parts that control the lips, the jaw, the larynx, the tongue, areas that were functionally

01:09:10.000 disconnected from her vocal tract because of this stroke in the brainstem, which connects the brain

01:09:15.680 to those muscles. We did the surgery about three weeks later. We started our research sessions with

01:09:23.280 her. We connected the cable. It's basically an HDMI cable that is attached to a head stage. The head stage

01:09:31.840 transforms the analog signals from her brain. These are a small voltage recordings.

01:09:37.760 I know I get into the weeds a bit much, but I think it's kind of interesting in signal processing.

01:09:42.720 Can you explain why the brain is analog and why you have to convert that to digital?

01:09:47.840 Well, to some degree, there's a level that it is digital. Like when we talk about-

01:09:52.480 Action potentials are digital.

01:09:53.600 Single neurons, action potentials, like firing yes or no, like a digital form. But when we're recording

01:10:00.960 a lot of these, especially at the ECOG, it's an analog. It's looking at the average of these from

01:10:06.560 a population of, let's say, a couple thousand of those neurons activating. And the work that we've done

01:10:12.480 actually over the last decade and a half, which led up to this, using methods like ECOG, we've

01:10:18.800 learned from that there's a map, what we alluded to in the very beginning, like the homunculus,

01:10:23.280 but a mini homunculus, that is corresponding to those parts of the vocal tract, the larynx,

01:10:30.720 the tongue, the jaw. We figured out essentially how those signals correspond to every consonant and

01:10:36.000 vowel in English about a decade ago. That was the impetus for actually starting this clinical trial,

01:10:43.840 was that we essentially had identified what the neural code was like, what part of the brain,

01:10:49.520 and how that neural activity corresponds to all the movements that create syllables, for example.

01:10:55.280 How big a data set was required to create that knowledge?

01:10:58.640 Probably about 36 participants' data to just get the basic idea.

01:11:05.600 And is this something where if that were 36,000, it would be how much better?

01:11:11.200 Hopefully perfect, like near perfect. We'll get to that because that's where things are going.

01:11:16.560 Okay.

01:11:17.120 How does this scale, how do we use the information from other individuals to help N1, for example.

01:11:23.840 But in Anne's particular case, we started from the beginning. We actually didn't use

01:11:28.160 that data. We knew that it was possible. We knew what the nature of that data and that code

01:11:33.520 would look like. And then at the same time that we're doing all of this research, Peter,

01:11:38.720 AI is developing in parallel. All of these tools that we now are using every day, transcribing our

01:11:47.200 voice right now into text, we use that technology. We can actually use those same technologies that generate

01:11:54.560 voices called speech synthesis. We've used a lot of the same tools, machine learning tools that are in

01:12:02.400 modern day AIs. We're now applying them on the brain activity and trying to use them to not translate,

01:12:10.240 for example, text and synthesized speech. But now the equation is different. It's translating from brain

01:12:16.640 activity to synthesized speech. The input is not text. The input is the ECOG activity across these 253 sensors.

01:12:26.480 Which of course is the logical extension. If you ignore the cost of compute, is there an advantage

01:12:33.600 to doing it that way because you take out an intermediary step?

01:12:36.800 Yes. It's because we know that the ballpark is there, but we know that everyone's brain

01:12:43.600 at that detail level, if you're going to reconstruct the words, you can't just be in the ballpark.

01:12:48.640 You have to know basically like what each leaf of grass on that ballpark is doing. And that's

01:12:55.600 highly variable across individuals. So what is AI? I'm just trying to think about

01:13:01.760 what the machine has to do. What is its training set? I'll walk you through. So the way that we

01:13:09.120 train the algorithm, the way that we started this was we would give an prompts on a screen

01:13:15.680 in text. And basically we would ask her to try to say it. Can she move her lips?

01:13:20.560 She can move a little bit, but she can't speak. So she can move her jaw, her lips,

01:13:24.000 but none of it is intelligible. She has what we call anarthria. Basically she can vocalize a 1.00

01:13:29.840 little bit, but none of it is intelligible in any form. I see. But if her inner monologue,

01:13:36.800 if you put up the word, the cow jumped over the fence and she says, and in her mind,

01:13:42.480 the signal is the cow jumped over the fence. Then I totally see how it works. At that point,

01:13:47.360 you have infinite training data. You would basically just have her read war and peace.

01:13:52.000 Right. So let me just clarify. The area that we are decoding from is not the part of the brain

01:13:57.440 that it's processing either inner monologue or reading. It really is this part that is about

01:14:04.480 this volitional intent. Oh, that's such a good point.

01:14:07.360 To speak, right? So it's not about her perception. It's not the reading.

01:14:11.360 So when I'm reading the cow jumped over the fence, if I just go,

01:14:16.240 what part of my brain is internalizing the cow jumped over the fence?

01:14:19.280 Well, your visual cortex. And then as it goes further, it's going into some of the language

01:14:24.240 areas, but it's not necessarily activating the lips, jaw, and the larynx, the areas that are

01:14:28.800 paralyzed. And so we're tapping into a part of the brain that is really-

01:14:33.040 So this is a hard exercise. This takes a lot of effort on her part. 1.00

01:14:36.720 A lot. So let me describe actually what that was. So for days, what we would do is have a sentence

01:14:45.200 on a screen and we'd give her the start time and the end time. And during that, she would just look

01:14:51.440 at the sentence. She'd give it a go cue and then try to say it. Nothing intelligible comes out.

01:14:57.120 She may or may not be moving the lips job, but just try to say, and that turns out to be very

01:15:01.520 important. Oh my God.

01:15:02.640 Like you can't just think about it. You can't just read it.

01:15:04.640 Yeah. Oh, wow.

01:15:05.600 You have to actually try to say it. And that's what she did. So we started with a very simple

01:15:12.480 vocabulary of about 27 words. The words that we chose are the NATO code words,

01:15:18.640 alpha, bravo, charlie, delta, echo. We did that because we could measure basically the accuracy

01:15:26.960 of the decoder that was analyzing those brain signals and translating them into those 26 different

01:15:33.440 code words. And on the first day, we were able to train the algorithm on a data set of maybe about

01:15:40.560 an hour and a half to get to about 50% accuracy. Does 50% accuracy mean she could get half of them

01:15:48.960 right? Or anytime you showed her one, there was a 50% chance it would be correct.

01:15:55.280 Both. What you just said is in our sense, identical.

01:15:58.480 Okay. But there wasn't a bias towards a subset of them that she was always getting right and others

01:16:02.960 that she was always getting wrong. Actually, there actually was a bias.

01:16:05.680 Yeah. Actually, if I get into details, yes. Some of them were more discriminable than other

01:16:10.480 words. And was it based on the number of syllables?

01:16:13.040 Yes. Actually, it was based on that in some of the phonetic properties. But one of the reasons why

01:16:18.720 NATO code words for us was a really useful training task for us is because NATO code words were developed

01:16:27.360 in the first place by the military to improve communication accuracy. The reason why we actually

01:16:33.840 use those code words is because sometimes if you just say A, B, D, Z, there's a lot of confusion.

01:16:41.120 So that's why we actually use those code words. It increases the discriminability and intelligibility

01:16:46.000 where a lot of those settings, you just can't make those errors. For example, pilots and the call

01:16:51.040 numbers, for example. So we use that because it has high discriminability. And on that first day,

01:16:58.160 I think we got about 50%.

01:16:59.760 This is going straight to voice? This is going to text?

01:17:02.160 This is straight to text.

01:17:03.120 Okay.

01:17:03.520 Yeah. And so we're just trying to figure out, could we decode which word it was? And it

01:17:08.880 was displayed in text.

01:17:09.920 That's the first day, 50%.

01:17:11.680 That's the first day.

01:17:12.720 Yeah. And then over the next, I would say about six days, the performance just got better and better.

01:17:19.760 And then by about like a week into this, she was up into the 95, 100% range. So that was unexpected.

01:17:28.080 It was incredible to see the performance increase so quickly, but that did take a full week.

01:17:33.920 Ask a question now. And I'm sure that the NATO code is not designed for this purpose, but presumably

01:17:41.280 one could concoct a series of words that contain within them the full range of tones, of phonetics,

01:17:52.640 of syllable juxtapositions that would allow you to use the smallest possible training data

01:18:00.240 to get the largest possible outcome. Does that make sense?

01:18:03.360 Absolutely.

01:18:04.400 How would one even develop such a thing? Because this is a novel problem.

01:18:07.760 Right. It's actually a really important and more profound actually than you may realize.

01:18:12.800 What you're referring to is the generative property of speech and language. And what I mean by generative

01:18:19.360 is that you can take these individual elements like consonants and vowels, which by themselves

01:18:23.760 have no meaning at all, and give rise to all possible meaning from just different combinations

01:18:30.240 of them, just like DNA. DNA, we've got four base pairs essentially as a code for all of life.

01:18:36.880 Except DNA is so much easier because it's finite and the rules are always the same.

01:18:45.280 You can define all the rules. Here you have, there's only four base pairs and they can only combine in

01:18:51.920 two ways. And everyone has a one-to-one mapping with what it's going to become. Here you have 26

01:18:58.080 letters. They can combine in a near infinite ways. And then there are all these dumb exceptions.

01:19:02.560 Right. So that's where the AI comes in. Let me just explain a little bit about how the algorithm works. 0.94

01:19:09.120 Because what you asked about actually is very, very much at the heart of the way that we do this.

01:19:14.480 So we don't go from the brain activity directly to speech and words and sentences. In the very

01:19:21.920 beginning with the NATO, that's what we do. You can use an algorithm called a classifier. It's going to

01:19:27.040 look at the pattern activity and then just say, okay, it looks mostly like beta. Another one looks

01:19:32.240 mostly like echo. Another one looks like Charlie. Okay. But to get to expressive, normal speech,

01:19:38.400 you need something that actually can open up much more combinatorial potential to generate sequences of

01:19:45.280 syllables, words, and sentences. So what we did was we built a decoder that translates the brain

01:19:53.440 activity patterns in very small segments, 10 to 20 millisecond little chunks of brain data,

01:20:02.400 really small signals, small windows of signals. And the machine learning is looking at those small

01:20:08.400 windows and making an educated guess. How does the mapping of that brain activity relate to a given

01:20:16.160 consonant or vowel? Now I'm using consonant and vowel just because it's easy to understand.

01:20:20.720 The reality is we used a speech unit. Like a phoneme or something like that?

01:20:26.000 A phoneme, yeah. But actually something that was statistically derived from a speech recognition

01:20:31.440 algorithm. It was statistically derived. It was not something that was linguistically or

01:20:36.720 that you read about. It's really a computational unit that we know if you can decode 100 of these

01:20:42.000 units, you can generate fluent, comprehensible speech. So we used AI actually to derive what those

01:20:48.560 units would be in the first place. We took a speech recognition system that Meta had made open source

01:20:53.680 about five years ago. It's one of the leading speech recognition algorithms. We took essentially

01:20:59.520 the neurons and what they do in that neural network and then we try to map those actually to the brain

01:21:05.680 activity patterns. That's on the very front end. The first step of the decoding, it's translating the

01:21:11.040 neural activity patterns to these individual speech units that are just 10 to 20 milliseconds long.

01:21:16.400 And then it of course knows the sequence of these units over time because it's part of the algorithm

01:21:22.560 calculation. And we use something called a language model, which is something that all of us are now

01:21:27.360 familiar with when you're texting and autocorrects your speech. Why? Because it has got a model of

01:21:31.600 English in there and that it knows what the particular sequence of the things should be like. And so even if

01:21:37.600 a lot of the data is kind of fuzzy, as more data accumulates, you get a sequence and then it can

01:21:42.640 basically use a best guess over time, what we call probabilistic inference of what was the most likely

01:21:50.400 word or phoneme at any given time point. And ultimately we could construct sentences.

01:21:55.520 Did you get a sense from Anne as to how her level of fatigue with this progressed? In other words,

01:22:04.000 what becomes the bottleneck? Does it get easier and easier for her to go through this talking motion

01:22:11.680 as she practices more? Is it just like any other muscle that we think of that has sort of atrophied

01:22:17.440 and now she's sort of getting her talking back in shape? 0.99

01:22:20.560 It is a bit of that. We're trying to make that easier over time. I think in the beginning days,

01:22:26.400 we're trying everything to get it to work. And a lot of it, again, has to do with this volitional

01:22:32.800 intent to speak. That turns out to be the most critical thing. One of the things that I thought

01:22:37.920 was really interesting also was we were doing so much decoding through these tasks that over time,

01:22:44.640 actually a couple months into this and it reported to us actually the strength of her orophacial

01:22:51.200 muscles, her jaw, the tongue, they were actually getting stronger through this constant therapy,

01:22:58.080 constant rehabilitation. And so I think right now everything is about just decoding the brain activity

01:23:04.240 to an artificial digital thing. But I do think that in the future, BCIs are also going to be a way that

01:23:11.040 we can do rehabilitation. It's a way that we have this direct readout of what the brain is trying to

01:23:17.040 do. You can essentially build a prosthetic that helps people speak, but in the process,

01:23:23.360 someone who hasn't spoken for a while will regain some of that natural strength over time. So that's

01:23:28.080 a new indication that we're thinking about in the future, how to use this technology actually to

01:23:31.920 augment and accelerate rehabilitation.

01:23:34.320 If Ann had that stroke today, how different, if at all, would this process look if you were

01:23:43.520 working with a person who hadn't spent 20 years or 18 years without speaking?

01:23:47.600 There's no question that I think it would work faster. There's less to learn. For her,

01:23:55.600 not speaking for 18 years basically meant that she basically had to relearn how to speak and we had to

01:24:03.680 keep up with her relearning. Her brain was probably reorganizing, relearning actually some of those

01:24:10.160 fundamental things. And she could see the feedback of essentially whether or not what she was trying

01:24:15.040 to say was right or wrong. And it was very intense work. So we're trying to make that easier over time.

01:24:20.560 But I think certainly the more preserved, the more recent that activity is, those memories,

01:24:27.440 the synapses we talked about earlier, the more stable, the more functional they are,

01:24:32.560 the easier it is to actually decode them.

01:24:34.960 So what will be the ceiling for the current technology? How many words per minute and at

01:24:42.000 what resolution or accuracy do you think the current technology, because this was ECOG in her case,

01:24:48.880 correct?

01:24:49.280 Right.

01:24:49.680 Where do you think it's going to go? Where will this asymptote?

01:24:52.880 We're seeing a lot of progress in this field. At the same time or soon after what we were doing,

01:24:59.360 there were other groups that basically could see similar effects. Ours was primarily from the

01:25:05.040 brain surface. Other groups, close colleagues of mine, were able to now do this with electrodes

01:25:10.560 that were inside the brain, you're seeing these intracortical arrays. So it seems that it's possible

01:25:16.080 possible with different approaches. I think what is going to be a key question is what's going to be

01:25:21.920 the right form moving in the future for many patients. With Ann, we were able to get about 80

01:25:28.240 words per minute on average. So sometimes much faster than that.

01:25:32.400 In comparison, how many words can you and I speak comfortably?

01:25:36.000 You and I are probably doing about 150, 160 words per minute right now.

01:25:40.000 Wow. So she could speak at half the rate you could speak at. That's pretty amazing.

01:25:44.800 Yeah. And it's not like the speech was coming out super slow. It's just that there's this built-in

01:25:50.240 latency time that we use to translate the brain activity into those words and sentences. And what

01:25:57.120 we published in 23...

01:25:57.360 You had a very short latency in your more recent paper, didn't you?

01:26:01.280 That's exactly right. In the 23 paper, our decoding strategy was to take this sequence of decoded

01:26:09.760 phonetic elements and we could look at that sequence and then apply the decoding algorithm in the language

01:26:15.920 model to reconstruct full sentences. And then we could even synthesize them, in fact, and personalize

01:26:21.840 them actually to her pre-injury voice. In a more recent study that we just published this year,

01:26:26.960 we were able to do this in a streaming way with less than a second latency between each phonetic

01:26:32.720 element. So it's not like we're waiting for the whole sentence to occur, but we're doing decoding

01:26:37.440 on the fly and it's intelligible and fast. And that will get the words up to what you think?

01:26:43.840 As quickly as she can try to say them, basically.

01:26:46.560 And this is all with the same hardware?

01:26:48.640 This is with the same hardware. Totally different algorithm.

01:26:51.920 On the intracranial hardware, obviously there's a big material science push to come up with the

01:26:57.840 most immunologically inert substance possible. That's your challenge there. But with the ECOG,

01:27:05.280 is there another hardware step function you're waiting for?

01:27:07.920 Not really. I mean, I think the thing that's most exciting about this is that

01:27:12.080 we have the technology now. We got to optimize it in the right form factor.

01:27:16.320 I mean, I guess it's just moving to a fully implantable device so you don't have to deal

01:27:20.800 with the infection risk. So we need to have the array that will have a lot more channels,

01:27:25.280 actually. So last time I talked about a credit card size with 253, we'd like to have something

01:27:31.120 that has 4X that amount of sensors. This seems completely achievable when you think

01:27:36.480 about what NVIDIA is doing or TSMC. I mean, that strikes me as very solvable.

01:27:41.440 It is. And we are doing it right now. With any medical device, you got to put it all together

01:27:45.680 and improve it. So we've taken these components that have very high bandwidth wireless connected to

01:27:52.560 this array. And I think in many ways we've done the hard part already, like what Ann did, what

01:27:58.720 Pancho did. He was one of our earlier participants. What Walter is doing. These are incredible people

01:28:04.160 that were really the first people in the world, actually, to be able to achieve this. Real pioneers.

01:28:09.520 That was the hard part. The hard part is always the first time.

01:28:12.800 Yeah, for sure. It's the proof of concept.

01:28:14.560 It's the proof of concept. Everything now is actually just about optimization, to be honest

01:28:18.800 with you. Do you think of this more as an engineering problem now?

01:28:21.360 It is. Yes.

01:28:23.040 Let's now expand it. So you have the proof of concept for the engineering problem that says brain works,

01:28:31.520 motor system doesn't work, we can extract speech. What about these other problems that we talked

01:28:38.960 about at the outset? What about ALS? Not for speech, but for respiratory function. A patient with ALS is,

01:28:47.840 I assume, I don't actually believe it or not know much about it, but I assume that they ultimately

01:28:52.240 succumb to respiratory complications and whether it be aspirations or things like that. So if we could

01:28:58.240 overcome that problem and bypass the degenerative motor neurons, is there an engineering solution

01:29:05.600 to ALS based on the type of technology we're seeing today?

01:29:09.760 When we say solution, I mean to preserve communication for someone?

01:29:14.000 Well, I would say let's go even beyond the ability to talk, but the ability to breathe normally,

01:29:19.680 for example, and ultimately the ability to not lose motor function outside of the CNS.

01:29:25.360 Yeah. So to do that basically is another couple of step functions in engineering where you basically

01:29:32.480 are talking about bypassing pretty significant section of the nervous system. So you're going

01:29:38.880 to tap into the brain to get some of the control signals. Some of this, you don't even need to tap

01:29:43.280 in the brain. For breathing, a lot of it is, as you know, is wired. We're not thinking about it,

01:29:48.160 certainly. Central pattern generators in the brainstem, for example, are really important for that

01:29:52.960 breathing pattern. This might sound naive, Eddie, but why is it that we couldn't wire into all of

01:29:58.640 the cranial nerves outside of the cranium and create a respiratory system that is fully automated?

01:30:04.880 Like almost think of an AICD for the diaphragm and chest wall. So we've spent almost all of our

01:30:11.360 time really talking about the brain side, but then you can imagine another, a whole new enterprise and

01:30:16.880 endeavor of building the electronics that not necessarily even tap into the nerves, the cranial

01:30:23.840 nerves, let's say, or the cervical nerves that go to the diaphragm, but you bypass those too and you

01:30:30.720 go directly to the muscles. Yeah.

01:30:32.480 Exactly. And so there is a field, we're not directly doing this research ourselves, but it's

01:30:37.840 highly related to where the future is. It's called FES, functional electrical stimulation.

01:30:42.480 So coupling the brain computer interface, the device that's decoding the brain activity,

01:30:49.280 translating into the control signals, and then actually acting on the muscles through stimulating

01:30:55.920 electrodes that are in the muscles themselves and doing that coordinated movement. Breathing

01:31:00.640 actually is a really interesting one because it's not as complex as like restoring our hand.

01:31:04.480 It's interesting. Everybody assumes today, if you really want to be in the forefront of technology,

01:31:09.120 you need to be on the CS side. But the truth of the matter is you need just as much horsepower on

01:31:14.640 the bioengineering side here, electrical engineering, biomedical engineering, mechanical engineering.

01:31:19.440 I mean, these are material science. I mean, these are, this type of problem is the intersection of

01:31:25.840 everything that is high tech from AI to computer science to all disciplines of engineering,

01:31:31.760 coupled with medicine. I mean, you have to have the surgeon too.

01:31:33.920 That's exactly right. And I think that you hit the nail on the head because in many ways,

01:31:38.400 that's the challenge, actually, more than the technology itself. It's really,

01:31:42.720 how do you get the engineer in the room with the neurosurgeon, with the neurologist,

01:31:46.160 the neuroscientist, all thinking in a really conservative way about solving this problem.

01:31:51.760 And then what you're going to see in the future, actually, is that this is going to evolve more

01:31:55.840 and more as a biological problem. Thinking about biology is the next technology solution,

01:32:01.760 engineered cells that are interfacing with the brain, as opposed to metal electrodes,

01:32:07.360 new ways of doing computing that are through biology, that are not through semiconductors.

01:32:12.560 That, I think, ultimately is where things are going to go in the future.

01:32:15.680 Well, say more about that. I mean, this is, there are some people that are already talking

01:32:18.960 about this, but I'd like people to understand more what you mean by that, because it's complicated.

01:32:22.800 It is complicated. What I'm talking about really is, I think, the next couple of steps. But one of the

01:32:29.040 reasons why this comes up is that you actually said it really precisely before. Okay, you've got

01:32:34.960 this electrode system. Let's say you're recording from one cell. Best case scenario, you'd be an

01:32:40.400 electronic system that can maybe do 10, maybe 40 in the future, a thousand channels. But the

01:32:46.480 denominators, 86 billion. We're not in the scale, not in the same regime of scale. Biology has done

01:32:53.680 that all along. Biology has solved a lot of these scaling problems. Cells that have the same genetic

01:33:00.080 programming multiply. Because of their environment, other factors, it becomes specialized for a specific

01:33:06.400 function. That's how our brain is. Each individual cell has the same genetic program, but because of its

01:33:13.280 local milieu, ends up having a different identity, different purpose. And so I think that is really

01:33:20.160 thinking outside of the electronical engineering, really moving into the realm of bioengineering.

01:33:26.720 And this field is moving pretty fast. There's a whole field that we call organoids. This is creating

01:33:32.800 mini-brains from cell cultures or stem cells, building miniature brains, primarily being used as models of

01:33:40.320 disease right now, but also as ways to test new drugs. But we're going to see these now interfacing

01:33:46.800 with the world of brain-computer interfaces. And so I think that that's part of the future,

01:33:52.720 for sure. It's very exciting. It's not near-term, but there certainly is something about the future

01:33:59.360 of technologies actually in biology. What is your stretch goal for the field in 2030? So stretch goal,

01:34:06.400 meaning I define that as things have to go well, but we're not talking science fiction.

01:34:11.600 By 2030, I hope that we have these systems actually available to a much broader market. Like

01:34:18.720 we have shown in a research setting, very controlled setting, that this can be done, the proof of concept.

01:34:25.760 What really needs to be done is a lot of hard engineering to make this practical, usable,

01:34:32.880 useful for people with a variety of different neurological conditions, not just ALS, but spinal

01:34:39.840 cord injury, stroke, multiple sclerosis. And that's a challenge. Everyone may have a very specific need.

01:34:47.120 We need to be able to solve that. It is an optimization. It can be solved. That's what I'd

01:34:51.760 love to see by 2030. Let's get a couple of these across the finish line so that they're actually out in

01:34:58.560 the world helping people. Is there a current company or set of companies that are the natural

01:35:04.480 owner to solving this problem based on their existing expertise? Or is what you're talking

01:35:10.800 about basically new companies that have to become capitalized and do this de novo? Who would be the

01:35:16.400 natural owner of this? I think it's both. So the most famous probably is Neuralink, Elon Musk's

01:35:22.400 company that has a very specific approach where you have a robot that is surgically inserting and

01:35:29.840 sewing electrodes into the brain and trying to record from that very finest resolution. And I think

01:35:36.080 there's a lot of progress with that, but also we've seen a lot of challenges. It's a really hard

01:35:41.920 technical problem to solve at that scale. There's a variety of other companies in that vein. One of the

01:35:47.440 things that we're working on is a highly customized ECOG approach because basically we already know

01:35:53.520 that it works from a lot of the work that we've done and we can make it a lot higher resolution

01:35:58.720 than we've done before and make it much safer with a fully implantable system. And then we're going to

01:36:03.840 see more and more over time that this is going to become less and less invasive. Just like we were

01:36:08.640 talking at the very beginning of our conversation, surgeries have become less invasive over time.

01:36:14.240 Brain-computer interfaces will become less invasive. We're at the very beginning of this

01:36:19.120 story. Getting the most amount of data right now is the most important with highly invasive

01:36:25.120 approaches. But I think as time goes on, we're going to back out from that invasiveness. That's

01:36:29.840 always how things evolve to make it more generalizable, easier and safer for people to do.

01:36:35.360 Now, when you say less invasive, do you think there will ever be a day when you can do this

01:36:39.200 off an EEG on the surface? Or do you think, no, it will be more like minimally invasive surgery to

01:36:45.040 open surgery where instead of a craniotomy, we're going to bore a single hole in there. We're going

01:36:49.280 to put a small tiny chip in through the dura implanted on there and we're done. The latter.

01:36:55.360 The resolution at the outside of the skull is probably never going to be good enough.

01:37:00.880 We're talking about a physics problem. I think a lot of people have tried to solve.

01:37:04.640 Batteries will never store energy nearly as well as hydrocarbons, full stop.

01:37:10.480 That level of resolution that we have from the scalp, in theory, I think, but in practice,

01:37:16.320 no one has been able to crack that. A lot of smart people have worked on that problem.

01:37:20.320 Interesting.

01:37:21.200 I do know that devices can continue to be miniaturized. I know that surgery

01:37:25.920 can continue to be safer. So we will see this point in history where

01:37:30.480 devices at some point are not going to just be about medical applications. They'll be

01:37:36.080 essentially enhancement level. There's huge ethical questions that we're going to have

01:37:40.480 to deal with when that time comes. We're not there right now. But I would bet on the technology.

01:37:46.960 We're not talking about breaking any rules and laws of physics in order to get there. We're just

01:37:51.360 talking about scaling electronic or miniaturizing it in a way that is just a smaller form factor. But

01:37:57.200 over time everything becomes less invasive. So I'm sure you get asked this question all the time,

01:38:02.240 but going back to the origin of Ann's story, so many people suffer brain injuries.

01:38:08.240 If you could wave a magic wand, you would just hope for some regeneration of the

01:38:13.600 injured portion of her brain. And my guess is in the case of Ann, the actual total volume of cells 0.90

01:38:20.640 that are damaged is quite small. It could be this half the size of your thumb, right? I mean,

01:38:24.560 it's a relatively small, but it just happened to be in the most precious part of real estate in her

01:38:28.720 entire body. So do we know, or do you have any point of view on the potential future of stem cell

01:38:38.080 like interventions for the purpose of regeneration specifically in the CNS?

01:38:43.440 Yeah. I mean, this is an area that I think got a lot of focus and attention maybe about 10 or 15

01:38:49.040 years ago. And I would say largely the results were pretty modest.

01:38:53.120 Yeah, at best.

01:38:54.560 Yeah, at best. It's coming back now because of a lot of cell-based therapies, organoids,

01:39:02.240 building miniature models of brains on cell cultures, basically. I think the first things

01:39:08.800 that we're going to see and where I am seeing some promise is very focal delivery in replacing

01:39:14.240 cells that have been lost in small targets of the brain. So back to Parkinson's disease,

01:39:19.440 where you've got degeneration of dopaminergic neurons and the substantia nigra, the goal is can

01:39:26.080 you replace and basically transplant some stem cells into that part of the brain?

01:39:31.600 Remind me why the cells in the substantia nigra, do we know what's killing them?

01:39:37.040 It could be multiple fault. It's partly genetic. There are certain genes that predispose to

01:39:41.920 degeneration there. There are certain environmental toxins that can cause the degeneration. And then

01:39:47.280 there's like a huge bucket. We still don't know what's causing that. But at the end of the day,

01:39:52.640 there is a degeneration of those very specialized cells. Most of the treatments are around dopamine

01:39:57.040 replacement medications. And how close do you think we are towards transplant?

01:40:02.880 It's already been done actually like 20, 30 years ago.

01:40:06.000 Oh, really? I wasn't aware.

01:40:06.640 Yeah. Using fetal grafts.

01:40:08.560 They just didn't take?

01:40:09.680 Some of them took. In fact, some patients got benefit from it. The side effects were also fairly

01:40:14.720 severe. What kind of side effects?

01:40:16.720 If you have too much dopamine, you can actually get dyskinesias. So hyper movement. So one of the

01:40:21.920 cardinal symptoms of Parkinson's. Hypo movement.

01:40:24.640 Yeah. Bradykinesia specifically, where you have slowed movements, slowed initiate movements as well.

01:40:30.960 But if you have cells that are just pumping out dopamine, they can also be putting out too much and

01:40:35.760 you get the opposite effect. So it's not as simple as just putting them in there. They actually have to

01:40:41.200 be tuned in the right way to put out the right levels. So there's a new generation of new therapies

01:40:47.760 that we're really interested in trialing at UCSF that are much better cell models, much better

01:40:54.720 control of dopamine that's involved. We have much better delivery systems.

01:40:58.400 Could you imagine that? Could you imagine engineering your way out of Parkinson's disease?

01:41:02.640 We're working on it.

01:41:03.920 What about synthetic cells where you completely get to control it? So again, you have the substrate

01:41:08.960 problem, but if it's truly a synthetic cell, then presumably it can make dopamine as well,

01:41:13.520 as opposed to an implantable slow leak dopamine that you've come up with some slick way to refill.

01:41:17.920 But what do you think is more likely the more pure engineering approach or the more biologic

01:41:23.600 transplant approach where you just try to tune it?

01:41:25.920 The near term, of course, is taking some cell cultures that are not purely synthesized. That's

01:41:31.120 still, I think is a huge goal outside of just brain. Like, can you generate a cell de novo

01:41:37.360 without some origins?

01:41:38.320 And does that require immune modulation?

01:41:40.240 Oh, absolutely.

01:41:41.040 So it's a full transplant.

01:41:42.560 Yeah. So a lot of these patients initially will be on immunosuppression for that. But that's also

01:41:49.360 improved a lot.

01:41:50.480 As immunosuppressive as if they had a kidney transplant or a liver or heart transplant?

01:41:55.120 Yes.

01:41:55.440 Wow.

01:41:56.000 Yeah. I think that's primarily right now the level of precaution. There is progress being made

01:42:01.200 in trying to make these things as least immunogenic as possible. That's where a lot of the engineering

01:42:07.040 actually is focused on is just make it the least immunogenic to avoid a rejection scenario. So I am

01:42:14.400 excited about that. And that's some of the biological engineering that I was talking about. Biotechnology or

01:42:20.000 the future of technology, really coming back to the biology, moving a little bit away from the

01:42:24.560 electrical engineering.

01:42:26.080 So in 15 years, in 2040, you're still going to be operating. You'll probably be in the final

01:42:33.040 decade or 15 years of your career. So by a surgeon's standards, plenty of work to do.

01:42:38.880 What do you think the world looks like in 2040? Which major problems that stand in front of you

01:42:44.560 today do you expect to fall and what will be the implications?

01:42:47.360 I think that the course that things are changing and how many things are being unlocked right now,

01:42:55.360 we're close. I think we're really getting close. Some of these things are not standard because of the

01:43:00.320 side effect profiles are too severe, but they can have therapeutic efficacy. We need to do that,

01:43:06.480 tuning this optimization. There's a lot of proof of concept out there. But like I alluded to earlier

01:43:12.640 before, 99% of the work is in the optimization in that engineering. I do think that now that we

01:43:19.120 understand what are the molecular and genetic drivers of a disease as devastating as glioblastoma,

01:43:25.440 we will have way more powerful tools that will hopefully make it a chronic condition as opposed to

01:43:31.920 a life, a death sentence in 18 months on average. That being said, with surgeries, we can get out

01:43:37.920 to years, many years. But a goal would be to make a chronic potentially cure by essentially attacking

01:43:45.840 the mechanisms. We now know the genes that are altered. We need to be able to turn on the immune

01:43:50.640 system to recognize huge amount of effort in trying to figure this out. I do think, and I'm very optimistic,

01:43:57.280 around neurodegenerative disorders. There's just so many promising things, including the cognitive

01:44:02.320 ones like Alzheimer's. I think earlier diagnosis and earlier treatment is going to be the first thing

01:44:07.440 where we're going to have the best effects. That is a really difficult one. But around Parkinson's,

01:44:12.640 where there's a focal problem, you can regenerate those cells. So you're more optimistic on the

01:44:17.600 movement disorders than you are the cognitive disorders. That's right. Partly it's because

01:44:24.960 the target in the cell loss is very focal. We can get cells through a surgery. When we're talking

01:44:30.000 about Alzheimer's, it's a bit trickier because it involves multiple systems in the brain simultaneously.

01:44:36.720 There are studies even using electrical stimulation in parts of the brain that are really important

01:44:40.800 for encoding memory. These things are promising, but I think for these really step functions and what

01:44:45.680 everyone wants is to either stall the disease or reverse it. It's going to take more time. But I do think

01:44:51.120 the early detection is going to be a game changer. A little off topic, but it's come up through the

01:44:56.560 story of Anne. Do you have a point of view on things that place people at risk for vestibular

01:45:02.080 artery dissections? For example, for whatever reason, whether it's just a wives' tale or not,

01:45:07.280 I've always been afraid of having anybody ever adjust my neck for fear of having a vestibular

01:45:12.640 artery dissection. Is there any truth to that as a risk? Are there other things that people should be

01:45:17.440 aware of given the low probability, but very, very high severity of such an injury?

01:45:23.600 It's not a wives' tale. It's actually statistically proven that certain kind of chiropractic movements

01:45:29.680 around the neck can cause an injury to the wall of the vertebral artery. And that term dissection

01:45:36.240 means that the wall of the artery has dissected. There's usually multiple different layers to that

01:45:42.400 vessel wall. And what happens with the dissection is the vessel is injured and then blood actually

01:45:47.600 starts splitting the wall of the artery more up until the point where it becomes occluded. And so

01:45:54.320 it's a very, very dangerous situation. And like you said, a critical part of the brain stem. So

01:46:00.960 generally we recommend not severe aggressive movements, but sometimes you can see it actually

01:46:06.320 around sports where you have a very high velocity movement around imposter around the neck.

01:46:12.400 And so those are the other cases where you can see it. That being said, this is very low

01:46:19.200 incidents, very low probability of happening. It's not at the level that you could really tell people

01:46:24.880 to avoid certain sports or anything like that. If we could bring Harvey Cushing back from the dead,

01:46:30.720 then you could have dinner with him tonight. What do you think he would say if he saw what was going

01:46:35.760 on in the field that he created? I think that there would be one part of him that is looking at some

01:46:42.480 of the surgeries that we do where we're still doing craniotomies and he would say that looks pretty

01:46:48.640 similar to what we did 150 years ago. I think that's part of his genius. The fact that we still do it means

01:46:55.040 that it still works and it's still safe, gets people through. A lot of that credit goes to Dr. Cushing.

01:47:02.160 But there will be things that I don't think he could have ever conceived. The way that we're

01:47:07.760 retrieving blood clots that are reversing strokes. What we're doing with brain computer interfaces,

01:47:15.360 decoding brain activity, the substrate of thought to replace communication for people who are

01:47:22.080 paralyzed, I think that that would have been very hard to really imagine back then. Primarily because

01:47:28.000 our knowledge was so limited and electronics was nowhere even close to being able to imagine what

01:47:33.680 could be done now. So a lot of what we're seeing actually relies on technology that has evolved like

01:47:40.560 artificial intelligence. A lot of the work that we did on decoding the brain just couldn't work.

01:47:45.040 Even though we had the hardware maybe 10 or 20 years ago, probably even earlier than that,

01:47:50.560 the decoding was not possible until this modern machine learning. These things are just accelerating

01:47:56.800 very, very fast right now. When I was a resident, I used to have this very famous picture on my wall

01:48:02.800 of the five physicians who were sort of the founding physicians at Hopkins. So of course you had

01:48:07.920 Halstead in surgery and Osler in medicine and I think Kelly was gynecology and then there was a

01:48:13.360 pathologist and of course Cushing was the understudy of Halstead before he left for Harvard. I honestly

01:48:20.240 think if you could bring all of them back to life today to see how much each of their fields had

01:48:27.200 progressed, I think that Cushing would be the one most blown away because, and maybe I'm wrong and some

01:48:35.600 historian will correct me, but I really think that what we've talked about today is to your point

01:48:41.600 unimaginable. So of course Osler would see medications that he never could have conceived

01:48:47.040 them, right? He could never conceive of a GLP-1 agonist and the profound effect it could have

01:48:53.120 on weight loss. He could never conceive at the time that there would be a medication that could eradicate

01:48:57.920 cholesterol, let alone an injection once every six months that could do it. He might have not even

01:49:03.840 conceived obesity back then. That's a good point. Although he was tasting urine, so he certainly knew

01:49:10.080 about diabetes. But yeah, I think the mental leap to where we are, although look, maybe the pathologist

01:49:16.400 would have never imagined the genomic sequencing that we could do of tumors today. Of course back then

01:49:21.120 it was all histology. So it is amazing to me how much medicine has changed in 100 years. Of course,

01:49:28.240 it doesn't take a leap to imagine that if we're still around as a species in 100 years,

01:49:32.960 the next 100 years is going to offer far bigger changes. Absolutely. I mean, the pace of acceleration

01:49:39.200 now is unprecedented. The underlying reason why I think Cushing would be the one that would be the

01:49:46.480 hardest to understand what's happening now is because we are talking about the brain. We are

01:49:51.760 talking about an organ system that we're just starting to fathom and put around our heads around

01:49:57.840 sort of the complexity. For the last 150 years, neurosurgery has really actually been about how do

01:50:05.120 you avoid injuring the brain? How do you take a tumor out of it? How do you deal with the plumbing,

01:50:10.800 which is the vascular system, the blood supply? But if you think about it, the biggest open-ended

01:50:17.600 questions are really being addressed right now in the coming decades. How does the brain itself work?

01:50:24.240 And then how do we tap into that to address a large variety of neurological and psychiatric

01:50:30.080 conditions? The history of neurosurgery was actually primarily about trying to avoid injury,

01:50:36.000 stay outside of the brain, etc. Now it's much more inward-looking, trying to understand actually

01:50:42.480 how the system works, how the organ works. And it's a super exciting time because every time we unlock

01:50:48.880 essentially a function of a certain part of the brain, there's a very high probability that there's

01:50:53.360 going to be a therapy either through a brain-computer interface or through a new biological approach.

01:50:58.960 Every time we unlock a new mechanism, there'll be something that we can do to treat it and that's

01:51:04.480 what the future is going to look like. One of my hidden agendas of this podcast

01:51:09.520 is to encourage as many young people as possible to go into medicine. And I understand that today,

01:51:14.080 medicine is not nearly as attractive a career as it was 20 years ago, 30 years ago, 50 years ago,

01:51:19.440 and that the best and the brightest are typically going elsewhere. But I think a podcast like this,

01:51:23.760 as are many of the podcasts I do with doctors, I really hope it showcases that we need the best and

01:51:29.360 the brightest to go into this. And again, this is not saying we don't need another brilliant person doing

01:51:33.920 AI or investment banking or law or wherever else the top people go, but there is really an opportunity

01:51:41.680 to bend the arc of civilization by choosing a career in medicine. And what you're doing, Eddie,

01:51:49.120 is really on the forefront of that, especially the way it combines all disciplines of science,

01:51:56.880 medicine, and technology. It's just, it's super exciting.

01:51:59.280 Thanks, Peter. Yeah, I'm really excited for that too.

01:52:02.000 Thanks for coming. I really appreciate this discussion.

01:52:04.320 Thanks for having me.

01:52:05.360 Thank you for listening to this week's episode of The Drive. Head over to peteratiamd.com

01:52:12.240 forward slash show notes. If you want to dig deeper into this episode, you can also find me on

01:52:18.480 YouTube, Instagram, and Twitter, all with the handle peteratiamd. You can also leave us review on Apple

01:52:24.960 podcasts or whatever podcast player you use. This podcast is for general informational purposes only

01:52:31.520 and does not constitute the practice of medicine, nursing, or other professional healthcare services,

01:52:36.000 including the giving of medical advice. No doctor patient relationship is formed.

01:52:41.360 The use of this information and the materials linked to this podcast is at the user's own risk.

01:52:47.040 The content on this podcast is not intended to be a substitute for professional medical advice,

01:52:51.760 diagnosis, or treatment. Users should not disregard or delay in obtaining medical advice from any

01:52:57.360 medical condition they have, and they should seek the assistance of their healthcare professionals

01:53:01.920 for any such conditions. Finally, I take all conflicts of interest very seriously. For all of my

01:53:07.840 disclosures and the companies I invest in or advise, please visit peteratiamd.com forward slash

01:53:14.720 about where I keep an up-to-date and active list of all disclosures.

The Peter Attia Drive - September 08, 2025

#363 ‒ A new frontier in neurosurgery： restoring brain function with brain-computer interfaces, advancing glioblastoma care, and new hope for devastating brain diseases ｜ Edward Chang, M.D.

Episode Stats

Summary

Transcript