An AI smartwatch that detects seizures
-
0:02 - 0:03This is Henry,
-
0:03 - 0:05a cute boy,
-
0:05 - 0:07and when Henry was three,
-
0:07 - 0:12his mom found him having
some febrile seizures. -
0:13 - 0:18Febrile seizures are seizures that occur
when you also have a fever, -
0:18 - 0:20and the doctor said,
-
0:20 - 0:23"Don't worry too much.
Kids usually outgrow these." -
0:24 - 0:27When he was four,
he had a convulsive seizure, -
0:28 - 0:31the kind that you lose
consciousness and shake -- -
0:31 - 0:34a generalized tonic-clonic seizure --
-
0:34 - 0:41and while the diagnosis of epilepsy
was in the mail, -
0:42 - 0:44Henry's mom went to get him
out of bed one morning, -
0:45 - 0:47and as she went in his room,
-
0:49 - 0:53she found his cold, lifeless body.
-
0:56 - 0:58Henry died of SUDEP,
-
0:58 - 1:00sudden unexpected death in epilepsy.
-
1:02 - 1:05I'm curious how many of you
have heard of SUDEP. -
1:06 - 1:10This is a very well-educated audience,
and I see only a few hands. -
1:10 - 1:14SUDEP is when an otherwise
healthy person with epilepsy -
1:14 - 1:18dies and they can't attribute it
to anything they can find in an autopsy. -
1:20 - 1:24There is a SUDEP
every seven to nine minutes. -
1:24 - 1:27That's on average two per TED Talk.
-
1:31 - 1:35Now, a normal brain
has electrical activity. -
1:35 - 1:37You can see some of the electrical waves
-
1:37 - 1:40coming out of this picture
of a brain here. -
1:40 - 1:44And these should look
like typical electrical activity -
1:44 - 1:46that an EEG could read on the surface.
-
1:46 - 1:51When you have a seizure,
it's a bit of unusual electrical activity, -
1:51 - 1:52and it can be focal.
-
1:52 - 1:55It can take place
in just a small part of your brain. -
1:55 - 1:58When that happens,
you might have a strange sensation. -
1:58 - 2:02Several could be happening
here in the audience right now, -
2:02 - 2:04and the person next to you
might not even know. -
2:04 - 2:08However, if you have a seizure
where that little brush fire spreads -
2:08 - 2:10like a forest fire over the brain,
-
2:10 - 2:12then it generalizes,
-
2:12 - 2:16and that generalized seizure
takes your consciousness away -
2:16 - 2:18and causes you to convulse.
-
2:19 - 2:23There are more SUDEPs
in the United States every year -
2:23 - 2:26than sudden infant death syndrome.
-
2:26 - 2:29Now, how many of you have heard
of sudden infant death syndrome? -
2:29 - 2:31Right? Pretty much every hand goes up.
-
2:31 - 2:33So what's going on here?
-
2:33 - 2:38Why is this so much more common
and yet people haven't heard of it? -
2:38 - 2:41And what can you do to prevent it?
-
2:41 - 2:43Well, there are two things,
scientifically shown, -
2:43 - 2:46that prevent or reduce the risk of SUDEP.
-
2:47 - 2:50The first is: "Follow
your doctor's instructions, -
2:50 - 2:51take your medications."
-
2:51 - 2:53Two-thirds of people who have epilepsy
-
2:53 - 2:55get it under control
with their medications. -
2:56 - 3:00The second thing that reduces
the risk of SUDEP is companionship. -
3:00 - 3:05It's having somebody there
at the time that you have a seizure. -
3:05 - 3:09Now, SUDEP, even though
most of you have never heard of it, -
3:09 - 3:14is actually the number two cause
of years of potential life lost -
3:14 - 3:17of all neurological disorders.
-
3:17 - 3:21The vertical axis is the number of deaths
-
3:22 - 3:25times the remaining life span,
-
3:25 - 3:28so higher is much worse impact.
-
3:29 - 3:31SUDEP, however, unlike these others,
-
3:32 - 3:37is something that people right here
could do something to push that down. -
3:38 - 3:45Now, what is Roz Picard, an AI researcher,
doing here telling you about SUDEP, right? -
3:45 - 3:47I'm not a neurologist.
-
3:47 - 3:52When I was working at the Media Lab
on measurement of emotion, -
3:52 - 3:55trying to make our machines
more intelligent about our emotions, -
3:55 - 3:57we started doing a lot of work
measuring stress. -
3:59 - 4:01We built lots of sensors
-
4:01 - 4:04that measured it
in lots of different ways. -
4:04 - 4:06But one of them in particular
-
4:06 - 4:10grew out of some of this very old work
with measuring sweaty palms -
4:10 - 4:12with an electrical signal.
-
4:12 - 4:14This is a signal of skin conductance
-
4:14 - 4:16that's known to go up
when you get nervous, -
4:16 - 4:19but it turns out it also goes up with
a lot of other interesting conditions. -
4:19 - 4:22But measuring it with wires on your hand
is really inconvenient. -
4:22 - 4:26So we invented a bunch of other ways
of doing this at the MIT Media Lab. -
4:26 - 4:28And with these wearables,
-
4:28 - 4:33we started to collect the first-ever
clinical quality data 24-7. -
4:34 - 4:36Here's a picture of what that looked like
-
4:36 - 4:43the first time an MIT student collected
skin conductance on the wrist 24-7. -
4:43 - 4:46Let's zoom in a little bit here.
-
4:46 - 4:49What you see is 24 hours
from left to right, -
4:49 - 4:51and here is two days of data.
-
4:51 - 4:53And first, what surprised us
-
4:53 - 4:57was sleep was the biggest
peak of the day. -
4:57 - 4:58Now, that sounds broken, right?
-
4:58 - 5:03You're calm when you're asleep,
so what's going on here? -
5:03 - 5:05Well, it turns out
that our physiology during sleep -
5:05 - 5:08is very different
than our physiology during wake, -
5:08 - 5:10and while there's still a bit of a mystery
-
5:10 - 5:14why these peaks are usually
the biggest of the day during sleep, -
5:14 - 5:17we now believe they're related
to memory consolidation -
5:17 - 5:19and memory formation during sleep.
-
5:20 - 5:23We also saw things
that were exactly what we expected. -
5:23 - 5:26When an MIT student
is working hard in the lab -
5:26 - 5:27or on homeworks,
-
5:27 - 5:32there is not only emotional stress,
but there's cognitive load, -
5:32 - 5:37and it turns out that cognitive load,
cognitive effort, mental engagement, -
5:37 - 5:39excitement about learning something --
-
5:39 - 5:41those things also make the signal go up.
-
5:43 - 5:47Unfortunately, to the embarrassment
of we MIT professors, -
5:47 - 5:48(Laughter)
-
5:48 - 5:52the low point every day
is classroom activity. -
5:53 - 5:55Now, I am just showing you
one person's data here, -
5:55 - 5:58but this, unfortunately,
is true in general. -
6:00 - 6:05This sweatband has inside it
a homebuilt skin-conductance sensor, -
6:05 - 6:10and one day, one of our undergrads
knocked on my door -
6:10 - 6:13right at the end of the December semester,
-
6:13 - 6:15and he said, "Professor Picard,
-
6:15 - 6:18can I please borrow
one of your wristband sensors? -
6:18 - 6:21My little brother has autism,
he can't talk, -
6:21 - 6:23and I want to see
what's stressing him out." -
6:24 - 6:27And I said, "Sure, in fact,
don't just take one, take two," -
6:27 - 6:30because they broke easily back then.
-
6:30 - 6:33So he took them home,
he put them on his little brother. -
6:33 - 6:36Now, I was back in MIT,
looking at the data on my laptop, -
6:36 - 6:39and the first day, I thought,
"Hmm, that's odd, -
6:39 - 6:42he put them on both wrists
instead of waiting for one to break. -
6:42 - 6:44OK, fine, don't follow my instructions."
-
6:44 - 6:46I'm glad he didn't.
-
6:46 - 6:50Second day -- chill.
Looked like classroom activity. -
6:50 - 6:51(Laughter)
-
6:51 - 6:53A few more days ahead.
-
6:53 - 6:58The next day, one wrist signal was flat
-
6:58 - 7:02and the other had
the biggest peak I've ever seen, -
7:02 - 7:05and I thought, "What's going on?
-
7:05 - 7:08We've stressed people out at MIT
every way imaginable. -
7:09 - 7:11I've never seen a peak this big."
-
7:12 - 7:14And it was only on one side.
-
7:14 - 7:18How can you be stressed on one side
of your body and not the other? -
7:18 - 7:20So I thought one or both sensors
must be broken. -
7:21 - 7:23Now, I'm an electroengineer by training,
-
7:23 - 7:26so I started a whole bunch of stuff
to try to debug this, -
7:26 - 7:28and long story short,
I could not reproduce this. -
7:29 - 7:32So I resorted to old-fashioned debugging.
-
7:32 - 7:35I called the student at home on vacation.
-
7:35 - 7:40"Hi, how's your little brother?
How's your Christmas? -
7:40 - 7:43Hey, do you have any idea
what happened to him?" -
7:43 - 7:45And I gave this particular date and time,
-
7:45 - 7:46and the data.
-
7:46 - 7:50And he said, "I don't know,
I'll check the diary." -
7:51 - 7:54Diary? An MIT student keeps a diary?
-
7:54 - 7:56So I waited and he came back.
-
7:56 - 7:57He had the exact date and time,
-
7:57 - 8:01and he says, "That was right before
he had a grand mal seizure." -
8:03 - 8:06Now, at the time, I didn't know
anything about epilepsy, -
8:06 - 8:08and did a bunch of research,
-
8:08 - 8:12realized that another student's dad
is chief of neurosurgery -
8:12 - 8:14at Children's Hospital Boston,
-
8:14 - 8:16screwed up my courage
and called Dr. Joe Madsen. -
8:16 - 8:18"Hi, Dr. Madsen,
my name's Rosalind Picard. -
8:18 - 8:22Is it possible somebody could have
-
8:22 - 8:27a huge sympathetic
nervous system surge" -- -
8:27 - 8:29that's what drives the skin conductance --
-
8:29 - 8:31"20 minutes before a seizure?"
-
8:32 - 8:34And he says, "Probably not."
-
8:36 - 8:37He says, "It's interesting.
-
8:37 - 8:40We've had people whose hair
stands on end on one arm -
8:40 - 8:4220 minutes before a seizure."
-
8:43 - 8:44And I'm like, "On one arm?"
-
8:44 - 8:47I didn't want to tell him that, initially,
-
8:47 - 8:49because I thought this was too ridiculous.
-
8:49 - 8:51He explained how this could
happen in the brain, -
8:51 - 8:53and he got interested.
I showed him the data. -
8:53 - 8:56We made a whole bunch more devices,
got them safety certified. -
8:56 - 8:5990 families were being
enrolled in a study, -
8:59 - 9:02all with children who were going
to be monitored 24-7 -
9:02 - 9:05with gold-standard EEG on their scalp
-
9:05 - 9:07for reading the brain activity,
-
9:07 - 9:09video to watch the behavior,
-
9:09 - 9:12electrocardiogram -- ECG --
and now EDA, electrodermal activity, -
9:12 - 9:15to see if there was
something in this periphery -
9:15 - 9:17that we could easily pick up,
related to a seizure. -
9:18 - 9:25We found, in 100 percent
of the first batch of grand mal seizures, -
9:25 - 9:28this whopper of responses
in the skin conductance. -
9:28 - 9:30The blue in the middle, the boy's sleep,
-
9:30 - 9:32is usually the biggest peak of the day.
-
9:32 - 9:36These three seizures you see here
are popping out of the forest -
9:36 - 9:38like redwood trees.
-
9:39 - 9:43Furthermore, when you couple
the skin conductance at the top -
9:43 - 9:46with the movement from the wrist
-
9:46 - 9:51and you get lots of data
and train machine learning and AI on it, -
9:51 - 9:56you can build an automated AI
that detects these patterns -
9:56 - 10:00much better than just
a shake detector can do. -
10:00 - 10:04So we realized that we needed
to get this out, -
10:04 - 10:07and with the PhD work of Ming-Zher Poh
-
10:07 - 10:10and later great improvements by Empatica,
-
10:10 - 10:14this has made progress and the seizure
detection is much more accurate. -
10:14 - 10:17But we also learned some other things
about SUDEP during this. -
10:17 - 10:20One thing we learned is that SUDEP,
-
10:20 - 10:23while it's rare after
a generalized tonic-clonic seizure, -
10:23 - 10:26that's when it's most likely
to happen -- after that type. -
10:26 - 10:29And when it happens,
it doesn't happen during the seizure, -
10:29 - 10:32and it doesn't usually happen
immediately afterwards, -
10:32 - 10:34but immediately afterwards,
-
10:34 - 10:37when the person just seems
very still and quiet, -
10:37 - 10:42they may go into another phase,
where the breathing stops, -
10:42 - 10:45and then after the breathing stops,
later the heart stops. -
10:45 - 10:47So there's some time
to get somebody there. -
10:48 - 10:53We also learned that there is a region
deep in the brain called the amygdala, -
10:53 - 10:56which we had been studying
in our emotion research a lot. -
10:56 - 10:58We have two amygdalas,
-
10:58 - 10:59and if you stimulate the right one,
-
10:59 - 11:02you get a big right
skin conductance response. -
11:02 - 11:06Now, you have to sign up right now
for a craniotomy to get this done, -
11:06 - 11:09not exactly something
we're going to volunteer to do, -
11:09 - 11:12but it causes a big right skin
conductance response. -
11:12 - 11:16Stimulate the left one, big left
skin conductance response on the palm. -
11:16 - 11:20And furthermore, when somebody
stimulates your amygdala -
11:20 - 11:23while you're sitting there
and you might just be working, -
11:23 - 11:25you don't show any signs of distress,
-
11:26 - 11:27but you stop breathing,
-
11:28 - 11:32and you don't start again
until somebody stimulates you. -
11:33 - 11:34"Hey, Roz, are you there?"
-
11:34 - 11:36And you open your mouth to talk.
-
11:37 - 11:39As you take that breath to speak,
-
11:39 - 11:41you start breathing again.
-
11:43 - 11:46So we had started with work on stress,
-
11:46 - 11:49which had enabled us
to build lots of sensors -
11:49 - 11:51that were gathering
high quality enough data -
11:51 - 11:54that we could leave the lab
and start to get this in the wild; -
11:54 - 11:57accidentally found a whopper
of a response with the seizure, -
11:57 - 12:00neurological activation that can cause
a much bigger response -
12:00 - 12:01than traditional stressors;
-
12:01 - 12:04lots of partnership with hospitals
and an epilepsy monitoring unit, -
12:04 - 12:06especially Children's Hospital Boston
-
12:06 - 12:07and the Brigham;
-
12:07 - 12:10and machine learning and AI on top of this
-
12:10 - 12:13to take and collect lots more data
-
12:13 - 12:15in service of trying
to understand these events -
12:15 - 12:17and if we could prevent SUDEP.
-
12:18 - 12:22This is now commercialized by Empatica,
-
12:22 - 12:24a start-up that I had
the privilege to cofound, -
12:25 - 12:29and the team there has done an amazing job
improving the technology -
12:29 - 12:31to make a very beautiful sensor
-
12:31 - 12:34that not only tells time and does steps
and sleep and all that good stuff, -
12:34 - 12:38but this is running real-time
AI and machine learning -
12:38 - 12:40to detect generalized
tonic-clonic seizures -
12:40 - 12:42and send an alert for help
-
12:42 - 12:46if I were to have a seizure
and lose consciousness. -
12:46 - 12:48This just got FDA-approved
-
12:48 - 12:53as the first smartwatch
to get approved in neurology. -
12:54 - 13:01(Applause)
-
13:03 - 13:06Now, the next slide is what made
my skin conductance go up. -
13:07 - 13:09One morning, I'm checking my email
-
13:09 - 13:11and I see a story from a mom
-
13:11 - 13:13who said she was in the shower,
-
13:13 - 13:15and her phone was
on the counter by the shower, -
13:15 - 13:17and it said her daughter
might need her help. -
13:18 - 13:22So she interrupts her shower and goes
running to her daughter's bedroom, -
13:22 - 13:25and she finds her daughter
facedown in bed, blue and not breathing. -
13:25 - 13:29She flips her over -- human stimulation --
-
13:29 - 13:32and her daughter takes a breath,
and another breath, -
13:32 - 13:36and her daughter turns pink and is fine.
-
13:38 - 13:41I think I turned white reading this email.
-
13:41 - 13:43My first response is,
"Oh no, it's not perfect. -
13:43 - 13:45The Bluetooth could break,
the battery could die. -
13:45 - 13:48All these things could go wrong.
Don't rely on this." -
13:48 - 13:51And she said, "It's OK.
I know no technology is perfect. -
13:51 - 13:54None of us can always
be there all the time. -
13:55 - 13:59But this, this device plus AI
-
13:59 - 14:02enabled me to get there in time
to save my daughter's life." -
14:06 - 14:08Now, I've been mentioning children,
-
14:08 - 14:14but SUDEP peaks, actually,
among people in their 20s, 30s and 40s, -
14:14 - 14:15and the next line I'm going to put up
-
14:15 - 14:18is probably going to make
some people uncomfortable, -
14:18 - 14:20but it's less uncomfortable
than we'll all be -
14:20 - 14:23if this list is extended
to somebody you know. -
14:24 - 14:27Could this happen to somebody you know?
-
14:27 - 14:30And the reason I bring up
this uncomfortable question -
14:30 - 14:35is because one in 26 of you
will have epilepsy at some point, -
14:35 - 14:37and from what I've been learning,
-
14:37 - 14:40people with epilepsy often don't tell
their friends and their neighbors -
14:41 - 14:42that they have it.
-
14:42 - 14:47So if you're willing to let them
use an AI or whatever -
14:47 - 14:51to summon you in a moment
of possible need, -
14:51 - 14:53if you would let them know that,
-
14:53 - 14:55you could make a difference in their life.
-
14:56 - 14:59Why do all this hard work to build AIs?
-
15:00 - 15:01A couple of reasons here:
-
15:01 - 15:03one is Natasha, the girl who lived,
-
15:04 - 15:06and her family wanted me
to tell you her name. -
15:07 - 15:09Another is her family
-
15:09 - 15:11and the wonderful people out there
-
15:11 - 15:14who want to be there to support people
who have conditions -
15:14 - 15:17that they've felt uncomfortable
in the past mentioning to others. -
15:18 - 15:20And the other reason is all of you,
-
15:20 - 15:25because we have the opportunity
to shape the future of AI. -
15:25 - 15:28We can actually change it,
-
15:28 - 15:30because we are the ones building it.
-
15:30 - 15:32So let's build AI
-
15:32 - 15:35that makes everybody's lives better.
-
15:36 - 15:37Thank you.
-
15:37 - 15:42(Applause)
- Title:
- An AI smartwatch that detects seizures
- Speaker:
- Rosalind Picard
- Description:
-
Every year worldwide, more than 50,000 otherwise healthy people with epilepsy suddenly die -- a condition known as SUDEP. These deaths may be largely preventable, says AI researcher Rosalind Picard. Learn how Picard helped develop a cutting-edge smartwatch that can detect epileptic seizures as they occur and alert nearby loved ones in time to help.
- Video Language:
- English
- Team:
- closed TED
- Project:
- TEDTalks
- Duration:
- 15:54
Brian Greene edited English subtitles for An AI smartwatch that detects seizures | ||
Brian Greene edited English subtitles for An AI smartwatch that detects seizures | ||
Oliver Friedman edited English subtitles for An AI smartwatch that detects seizures | ||
Oliver Friedman edited English subtitles for An AI smartwatch that detects seizures | ||
Oliver Friedman edited English subtitles for An AI smartwatch that detects seizures | ||
Oliver Friedman edited English subtitles for An AI smartwatch that detects seizures | ||
Oliver Friedman approved English subtitles for An AI smartwatch that detects seizures | ||
Oliver Friedman edited English subtitles for An AI smartwatch that detects seizures |