YouTube

Got a YouTube account?

New: enable viewer-created translations and captions on your YouTube channel!

English subtitles

← An AI smartwatch that detects seizures

Get Embed Code
22 Languages

Showing Revision 13 created 04/25/2019 by Brian Greene.

  1. This is Henry,
  2. a cute boy,
  3. and when Henry was three,
  4. his mom found him having
    some febrile seizures.
  5. Febrile seizures are seizures that occur
    when you also have a fever,
  6. and the doctor said,
  7. "Don't worry too much.
    Kids usually outgrow these."
  8. When he was four,
    he had a convulsive seizure,
  9. the kind that you lose
    consciousness and shake --
  10. a generalized tonic-clonic seizure --
  11. and while the diagnosis of epilepsy
    was in the mail,
  12. Henry's mom went to get him
    out of bed one morning,
  13. and as she went in his room,
  14. she found his cold, lifeless body.
  15. Henry died of SUDEP,

  16. sudden unexpected death in epilepsy.
  17. I'm curious how many of you
    have heard of SUDEP.
  18. This is a very well-educated audience,
    and I see only a few hands.
  19. SUDEP is when an otherwise
    healthy person with epilepsy
  20. dies and they can't attribute it
    to anything they can find in an autopsy.
  21. There is a SUDEP
    every seven to nine minutes.
  22. That's on average two per TED Talk.
  23. Now, a normal brain
    has electrical activity.
  24. You can see some of the electrical waves
  25. coming out of this picture
    of a brain here.
  26. And these should look
    like typical electrical activity
  27. that an EEG could read on the surface.
  28. When you have a seizure,
    it's a bit of unusual electrical activity,
  29. and it can be focal.
  30. It can take place
    in just a small part of your brain.
  31. When that happens,
    you might have a strange sensation.
  32. Several could be happening
    here in the audience right now,
  33. and the person next to you
    might not even know.
  34. However, if you have a seizure
    where that little brush fire spreads
  35. like a forest fire over the brain,
  36. then it generalizes,
  37. and that generalized seizure
    takes your consciousness away
  38. and causes you to convulse.
  39. There are more SUDEPs
    in the United States every year

  40. than sudden infant death syndrome.
  41. Now, how many of you have heard
    of sudden infant death syndrome?
  42. Right? Pretty much every hand goes up.
  43. So what's going on here?
  44. Why is this so much more common
    and yet people haven't heard of it?
  45. And what can you do to prevent it?
  46. Well, there are two things,
    scientifically shown,
  47. that prevent or reduce the risk of SUDEP.
  48. The first is: "Follow
    your doctor's instructions,
  49. take your medications."
  50. Two-thirds of people who have epilepsy
  51. get it under control
    with their medications.
  52. The second thing that reduces
    the risk of SUDEP is companionship.
  53. It's having somebody there
    at the time that you have a seizure.
  54. Now, SUDEP, even though
    most of you have never heard of it,
  55. is actually the number two cause
    of years of potential life lost
  56. of all neurological disorders.
  57. The vertical axis is the number of deaths
  58. times the remaining life span,
  59. so higher is much worse impact.
  60. SUDEP, however, unlike these others,
  61. is something that people right here
    could do something to push that down.
  62. Now, what is Roz Picard, an AI researcher,
    doing here telling you about SUDEP, right?

  63. I'm not a neurologist.
  64. When I was working at the Media Lab
    on measurement of emotion,
  65. trying to make our machines
    more intelligent about our emotions,
  66. we started doing a lot of work
    measuring stress.
  67. We built lots of sensors
  68. that measured it
    in lots of different ways.
  69. But one of them in particular
  70. grew out of some of this very old work
    with measuring sweaty palms
  71. with an electrical signal.
  72. This is a signal of skin conductance
  73. that's known to go up
    when you get nervous,
  74. but it turns out it also goes up with
    a lot of other interesting conditions.
  75. But measuring it with wires on your hand
    is really inconvenient.
  76. So we invented a bunch of other ways
    of doing this at the MIT Media Lab.
  77. And with these wearables,
  78. we started to collect the first-ever
    clinical quality data 24-7.
  79. Here's a picture of what that looked like
  80. the first time an MIT student collected
    skin conductance on the wrist 24-7.
  81. Let's zoom in a little bit here.
  82. What you see is 24 hours
    from left to right,
  83. and here is two days of data.
  84. And first, what surprised us
  85. was sleep was the biggest
    peak of the day.
  86. Now, that sounds broken, right?
  87. You're calm when you're asleep,
    so what's going on here?
  88. Well, it turns out
    that our physiology during sleep

  89. is very different
    than our physiology during wake,
  90. and while there's still a bit of a mystery
  91. why these peaks are usually
    the biggest of the day during sleep,
  92. we now believe they're related
    to memory consolidation
  93. and memory formation during sleep.
  94. We also saw things
    that were exactly what we expected.

  95. When an MIT student
    is working hard in the lab
  96. or on homeworks,
  97. there is not only emotional stress,
    but there's cognitive load,
  98. and it turns out that cognitive load,
    cognitive effort, mental engagement,
  99. excitement about learning something --
  100. those things also make the signal go up.
  101. Unfortunately, to the embarrassment
    of we MIT professors,

  102. (Laughter)

  103. the low point every day
    is classroom activity.

  104. Now, I am just showing you
    one person's data here,
  105. but this, unfortunately,
    is true in general.
  106. This sweatband has inside it
    a homebuilt skin-conductance sensor,

  107. and one day, one of our undergrads
    knocked on my door
  108. right at the end of the December semester,
  109. and he said, "Professor Picard,
  110. can I please borrow
    one of your wristband sensors?
  111. My little brother has autism,
    he can't talk,
  112. and I want to see
    what's stressing him out."
  113. And I said, "Sure, in fact,
    don't just take one, take two,"
  114. because they broke easily back then.
  115. So he took them home,
    he put them on his little brother.
  116. Now, I was back in MIT,
    looking at the data on my laptop,
  117. and the first day, I thought,
    "Hmm, that's odd,
  118. he put them on both wrists
    instead of waiting for one to break.
  119. OK, fine, don't follow my instructions."
  120. I'm glad he didn't.
  121. Second day -- chill.
    Looked like classroom activity.
  122. (Laughter)

  123. A few more days ahead.

  124. The next day, one wrist signal was flat
  125. and the other had
    the biggest peak I've ever seen,
  126. and I thought, "What's going on?
  127. We've stressed people out at MIT
    every way imaginable.
  128. I've never seen a peak this big."
  129. And it was only on one side.
  130. How can you be stressed on one side
    of your body and not the other?
  131. So I thought one or both sensors
    must be broken.
  132. Now, I'm an electroengineer by training,
  133. so I started a whole bunch of stuff
    to try to debug this,
  134. and long story short,
    I could not reproduce this.
  135. So I resorted to old-fashioned debugging.

  136. I called the student at home on vacation.
  137. "Hi, how's your little brother?
    How's your Christmas?
  138. Hey, do you have any idea
    what happened to him?"
  139. And I gave this particular date and time,
  140. and the data.
  141. And he said, "I don't know,
    I'll check the diary."
  142. Diary? An MIT student keeps a diary?

  143. So I waited and he came back.
  144. He had the exact date and time,
  145. and he says, "That was right before
    he had a grand mal seizure."
  146. Now, at the time, I didn't know
    anything about epilepsy,

  147. and did a bunch of research,
  148. realized that another student's dad
    is chief of neurosurgery
  149. at Children's Hospital Boston,
  150. screwed up my courage
    and called Dr. Joe Madsen.
  151. "Hi, Dr. Madsen,
    my name's Rosalind Picard.

  152. Is it possible somebody could have
  153. a huge sympathetic
    nervous system surge" --
  154. that's what drives the skin conductance --
  155. "20 minutes before a seizure?"
  156. And he says, "Probably not."
  157. He says, "It's interesting.
  158. We've had people whose hair
    stands on end on one arm
  159. 20 minutes before a seizure."
  160. And I'm like, "On one arm?"
  161. I didn't want to tell him that, initially,
  162. because I thought this was too ridiculous.
  163. He explained how this could
    happen in the brain,

  164. and he got interested.
    I showed him the data.
  165. We made a whole bunch more devices,
    got them safety certified.
  166. 90 families were being
    enrolled in a study,
  167. all with children who were going
    to be monitored 24-7
  168. with gold-standard EEG on their scalp
  169. for reading the brain activity,
  170. video to watch the behavior,
  171. electrocardiogram -- ECG --
    and now EDA, electrodermal activity,
  172. to see if there was
    something in this periphery
  173. that we could easily pick up,
    related to a seizure.
  174. We found, in 100 percent
    of the first batch of grand mal seizures,

  175. this whopper of responses
    in the skin conductance.
  176. The blue in the middle, the boy's sleep,
  177. is usually the biggest peak of the day.
  178. These three seizures you see here
    are popping out of the forest
  179. like redwood trees.
  180. Furthermore, when you couple
    the skin conductance at the top
  181. with the movement from the wrist
  182. and you get lots of data
    and train machine learning and AI on it,
  183. you can build an automated AI
    that detects these patterns
  184. much better than just
    a shake detector can do.
  185. So we realized that we needed
    to get this out,
  186. and with the PhD work of Ming-Zher Poh
  187. and later great improvements by Empatica,
  188. this has made progress and the seizure
    detection is much more accurate.
  189. But we also learned some other things
    about SUDEP during this.

  190. One thing we learned is that SUDEP,
  191. while it's rare after
    a generalized tonic-clonic seizure,
  192. that's when it's most likely
    to happen -- after that type.
  193. And when it happens,
    it doesn't happen during the seizure,
  194. and it doesn't usually happen
    immediately afterwards,
  195. but immediately afterwards,
  196. when the person just seems
    very still and quiet,
  197. they may go into another phase,
    where the breathing stops,
  198. and then after the breathing stops,
    later the heart stops.
  199. So there's some time
    to get somebody there.
  200. We also learned that there is a region
    deep in the brain called the amygdala,
  201. which we had been studying
    in our emotion research a lot.
  202. We have two amygdalas,
  203. and if you stimulate the right one,
  204. you get a big right
    skin conductance response.
  205. Now, you have to sign up right now
    for a craniotomy to get this done,
  206. not exactly something
    we're going to volunteer to do,
  207. but it causes a big right skin
    conductance response.
  208. Stimulate the left one, big left
    skin conductance response on the palm.
  209. And furthermore, when somebody
    stimulates your amygdala
  210. while you're sitting there
    and you might just be working,
  211. you don't show any signs of distress,
  212. but you stop breathing,
  213. and you don't start again
    until somebody stimulates you.
  214. "Hey, Roz, are you there?"
  215. And you open your mouth to talk.
  216. As you take that breath to speak,
  217. you start breathing again.
  218. So we had started with work on stress,

  219. which had enabled us
    to build lots of sensors
  220. that were gathering
    high quality enough data
  221. that we could leave the lab
    and start to get this in the wild;
  222. accidentally found a whopper
    of a response with the seizure,
  223. neurological activation that can cause
    a much bigger response
  224. than traditional stressors;
  225. lots of partnership with hospitals
    and an epilepsy monitoring unit,
  226. especially Children's Hospital Boston
  227. and the Brigham;
  228. and machine learning and AI on top of this
  229. to take and collect lots more data
  230. in service of trying
    to understand these events
  231. and if we could prevent SUDEP.
  232. This is now commercialized by Empatica,

  233. a start-up that I had
    the privilege to cofound,
  234. and the team there has done an amazing job
    improving the technology
  235. to make a very beautiful sensor
  236. that not only tells time and does steps
    and sleep and all that good stuff,
  237. but this is running real-time
    AI and machine learning
  238. to detect generalized
    tonic-clonic seizures
  239. and send an alert for help
  240. if I were to have a seizure
    and lose consciousness.
  241. This just got FDA-approved
  242. as the first smartwatch
    to get approved in neurology.
  243. (Applause)

  244. Now, the next slide is what made
    my skin conductance go up.

  245. One morning, I'm checking my email
  246. and I see a story from a mom
  247. who said she was in the shower,
  248. and her phone was
    on the counter by the shower,
  249. and it said her daughter
    might need her help.
  250. So she interrupts her shower and goes
    running to her daughter's bedroom,
  251. and she finds her daughter
    facedown in bed, blue and not breathing.
  252. She flips her over -- human stimulation --
  253. and her daughter takes a breath,
    and another breath,
  254. and her daughter turns pink and is fine.
  255. I think I turned white reading this email.

  256. My first response is,
    "Oh no, it's not perfect.
  257. The Bluetooth could break,
    the battery could die.
  258. All these things could go wrong.
    Don't rely on this."
  259. And she said, "It's OK.
    I know no technology is perfect.
  260. None of us can always
    be there all the time.
  261. But this, this device plus AI
  262. enabled me to get there in time
    to save my daughter's life."
  263. Now, I've been mentioning children,

  264. but SUDEP peaks, actually,
    among people in their 20s, 30s and 40s,
  265. and the next line I'm going to put up
  266. is probably going to make
    some people uncomfortable,
  267. but it's less uncomfortable
    than we'll all be
  268. if this list is extended
    to somebody you know.
  269. Could this happen to somebody you know?
  270. And the reason I bring up
    this uncomfortable question
  271. is because one in 26 of you
    will have epilepsy at some point,
  272. and from what I've been learning,
  273. people with epilepsy often don't tell
    their friends and their neighbors
  274. that they have it.
  275. So if you're willing to let them
    use an AI or whatever
  276. to summon you in a moment
    of possible need,
  277. if you would let them know that,
  278. you could make a difference in their life.
  279. Why do all this hard work to build AIs?

  280. A couple of reasons here:
  281. one is Natasha, the girl who lived,
  282. and her family wanted me
    to tell you her name.
  283. Another is her family
  284. and the wonderful people out there
  285. who want to be there to support people
    who have conditions
  286. that they've felt uncomfortable
    in the past mentioning to others.
  287. And the other reason is all of you,
  288. because we have the opportunity
    to shape the future of AI.
  289. We can actually change it,
  290. because we are the ones building it.
  291. So let's build AI

  292. that makes everybody's lives better.
  293. Thank you.

  294. (Applause)