ABOUT THE SPEAKER
Dina Katabi - Technologist
Dina Katabi investigates how AI can make wireless devices sense human motion and vital signs.

Why you should listen

Dina Katabi designs new wireless devices that use machine learning to sense people through walls and occlusions. Her devices look like a Wi-Fi box. They transmit a low-power wireless signal and capture its reflections as it bounces off people and objects. They analyze those reflections to learn how people walk, measure their gait and detect elderly falls. The device can also measure a person's breathing, heart rate and sleep quality using wireless signals, without any sensor on the person's body. Katabi is working with medical doctors to use her technology to detect health emergencies and provide a better understanding of chronic diseases such as Alzheimer's and Parkinson's. 

Katabi is the Andrew & Erna Viterbi Professor of Electrical Engineering and Computer Science at MIT. She is also the director of the MIT's Center for Wireless Networks and Mobile Computing, a member of the National Academy of Engineering and a recipient of the MacArthur Fellowship. Her research has been recognized by the ACM Prize in Computing, the ACM Grace Murray Hopper Award, the SIGCOMM Test-of-Time Award, the IEEE William R. Bennett prize, the Faculty Research Innovation Fellowship, a Sloan Fellowship and multiple best paper awards. Several startups have been spun out of her lab, such as PiCharging and Emerald.

More profile about the speaker
Dina Katabi | Speaker | TED.com
TED2018

Dina Katabi: A new way to monitor vital signs (that can see through walls)

Filmed:
1,513,020 views

At MIT, Dina Katabi and her team are working on a bold new way to monitor patients' vital signs in a hospital (or even at home), without wearables or bulky, beeping devices. Bonus: it can see through walls. In a mind-blowing talk and demo, Katabi previews a system that captures the reflections of signals like Wi-Fi as they bounce off people, creating a reliable record of vitals for healthcare workers and patients. And in a brief Q&A with TED curator Helen Walters, Katabi discusses safeguards being put in place to prevent people from using this tech to monitor somebody without their consent.
- Technologist
Dina Katabi investigates how AI can make wireless devices sense human motion and vital signs. Full bio

Double-click the English transcript below to play the video.

00:13
When I was a kid,
0
1150
1175
00:14
I was, like many of you in this room,
very much fascinated by Star Wars,
1
2349
4900
00:19
and what fascinated me the most
is this notion of the Force,
2
7273
4476
00:23
this energy that connects
all people and all objects
3
11773
3110
00:26
and allows you to feel people
that you can't even see.
4
14907
3418
00:30
And I remember many nights,
I would be sitting at home,
5
18753
3088
00:33
just, like, concentrating and focusing,
trying to feel the Force,
6
21865
3970
00:38
and I didn't feel anything, don't worry.
7
26624
2563
00:41
(Laughter)
8
29211
1857
00:43
And later in life, I became a scientist.
9
31092
2850
00:45
I joined the MIT faculty
and started working on wireless signals.
10
33966
4713
00:51
These are things like Wi-Fi
or cellular systems,
11
39184
3143
00:54
and I did a lot of work in that domain.
12
42351
2352
00:57
But then, again, this Force thing
kept nagging me,
13
45489
4589
01:02
and at some point, I was just like,
14
50102
1925
01:04
"Wait a minute, these wireless signals --
they are like the Force."
15
52051
4652
01:08
So if you think about it,
16
56727
1194
01:09
wireless signals,
they travel through space,
17
57945
2631
01:12
they go through obstacles
and walls and occlusions,
18
60600
3480
01:16
and some of them,
19
64104
1241
01:17
they reflect off our bodies,
because our bodies are full of water,
20
65369
4150
01:21
and some of these minute reflections,
21
69543
2206
01:23
they come back.
22
71773
1258
01:25
And if, just if, I had a device that can
just sense these minute reflections,
23
73055
6971
01:32
then I would be able to feel people
that I cannot see.
24
80050
2984
01:36
So I started working with my students
on building such a device,
25
84444
3753
01:40
and I want to show you
some of our early results.
26
88221
3310
01:44
So here, you see my student standing,
27
92584
2283
01:46
and here is our device.
28
94891
1977
01:48
And we are going to put the device
in the other office, behind the wall,
29
96892
5610
01:54
and we are going
to monitor him as he moves.
30
102526
2817
01:57
This red dot is tracking him
using wireless signals.
31
105367
4079
02:02
And as you can see, the red dot
is tracking his movements very accurately,
32
110760
4784
02:07
purely based on how his body interacts
with the surrounding wireless signals.
33
115568
4754
02:13
Pretty accurate, isn't it?
34
121631
1704
02:16
He has no wearables, nothing.
35
124702
1873
02:18
(Applause)
36
126599
2814
02:21
Now you might be wondering,
37
129437
1745
02:23
how is it possible
that we can sense people
38
131206
3182
02:26
and track them, without
any wearables, through walls,
39
134412
3846
02:30
and the easiest analogy
to think about is radar.
40
138282
3366
02:33
I'm sure many of you
have seen this picture.
41
141672
2093
02:35
You transmit a wireless signal to the sky,
42
143789
2158
02:37
it reflects off some airplane,
comes back to you,
43
145971
2968
02:40
and you start detecting these airplanes.
44
148963
2341
02:44
But if it were just radar,
45
152328
1944
02:46
then we would have this 50 years ago.
46
154296
2253
02:49
So it's not just radar.
47
157272
1706
02:51
There are two key differences.
48
159285
2190
02:54
So the first difference, of course --
49
162229
1763
02:56
you can't, like radar, just blast
wireless power at somebody.
50
164016
3519
02:59
You're going to fry them
like if they were in a microwave.
51
167559
2727
03:02
Don't do that.
52
170310
1150
03:04
So it means that you have to be able
to deal with very weak signals,
53
172595
5065
03:09
and that means that your device
has to be very sensitive.
54
177684
3126
03:14
The second difference is that,
unlike the sky, where it's empty --
55
182144
4131
03:18
if you are lucky, there is one airplane
that you can catch there.
56
186299
3144
03:21
Like, look at the room
57
189467
1342
03:22
and look how many objects
and people there are.
58
190833
2888
03:25
So in indoor environments, the signal
not only reflects off the person,
59
193745
3937
03:29
if reflects off the person,
off the floor, the ceiling,
60
197706
3793
03:33
off other people around,
61
201523
2222
03:35
and you get very complex reflections
62
203769
1983
03:37
where the same signal reflects
off me and then off you,
63
205776
2731
03:40
and then off the ceiling,
then off the floor.
64
208531
2238
03:42
And you have to make sense of that mess.
65
210793
3169
03:47
But we were lucky.
66
215993
1849
03:49
We were coming at the right time.
67
217866
2492
03:53
So two things helped us.
68
221252
1872
03:55
The first thing is radiotechnologies
have evolved a lot,
69
223148
4056
03:59
and over the last decade,
70
227228
1775
04:01
radio technology
became much more powerful,
71
229027
2310
04:03
so we were able to build
very sensitive radios
72
231361
3737
04:07
that can sense weak and minute RF signals.
73
235122
3357
04:11
The second thing: machine learning.
74
239725
3039
04:14
So you keep hearing about machine learning
75
242788
2000
04:16
and there was a revolution
of machine learning recently,
76
244812
2961
04:19
in deep learning,
77
247797
1151
04:20
and that allowed us to build
machine-learning models
78
248972
3769
04:24
that can understand wireless signals
and interpret them
79
252765
3547
04:28
so they would know what happened
in the environment.
80
256336
3505
04:31
So if you think of it,
the radio is like the ear of our device
81
259865
3810
04:35
and the machine learning
is like the brain,
82
263699
2579
04:38
and together, they have
a very powerful device.
83
266302
3823
04:44
So what else can we sense about people
using wireless signals?
84
272108
5023
04:50
Sleep.
85
278458
1151
04:51
Sleep, actually, is something
very dear to my heart,
86
279633
2429
04:54
because my sleep is a disaster.
87
282086
1927
04:56
(Laughter)
88
284037
1348
04:57
So one thing is when you start working
on some physiological signal
89
285409
3229
05:00
and you discover that yours sucks.
90
288662
2659
05:03
(Laughter)
91
291908
3095
05:07
So you can see why we can capture sleep,
92
295344
2126
05:09
because the person walks and the device
sees him as he walks to bed,
93
297494
3266
05:12
when he stops tossing around in bed,
94
300784
2275
05:15
when he steps out of bed,
95
303083
1841
05:16
and that measure of sleep
is what people call actigraphy.
96
304948
3698
05:20
It's based on motion.
97
308670
1254
05:23
But it turned out
that we can actually get sleep
98
311087
3294
05:26
at a much more important level.
99
314405
2387
05:28
We can understand
the change in the brain waves
100
316816
3564
05:32
that occur during sleep.
101
320404
1548
05:35
So, many of you probably know
that as we go to sleep,
102
323403
2953
05:38
our brainwaves change
and we enter different stages:
103
326380
3373
05:41
awake, light sleep, deep sleep
and REM, or rapid eye movement.
104
329777
4069
05:46
These stages are of course
related to sleep disorders,
105
334267
4334
05:50
but they are also related
to various diseases.
106
338625
3110
05:53
So for example, disturbances in REM
are associated with depression.
107
341759
5461
05:59
Disturbances in deep sleep
are associated with Alzheimer's.
108
347718
3928
06:04
So if you want to get sleep staging,
109
352614
2539
06:07
today, you will send the person
to the hospital,
110
355177
2310
06:09
they put all of these
electrodes on their head,
111
357511
2554
06:12
and they ask them to sleep like that.
112
360089
1830
06:13
(Laughter)
113
361943
2060
06:16
It's not really a happy experience.
114
364027
2935
06:20
So what if I tell you
that I can do the same thing
115
368155
2936
06:23
but without any of these electrodes
on the person's body?
116
371115
3167
06:27
So here is our device,
117
375596
1872
06:29
transmitting very low power
wireless signal,
118
377492
2762
06:32
analyzes the reflections using AI
119
380278
2112
06:34
and spits out the sleep stages
throughout the night.
120
382414
3166
06:38
So we know, for example,
when this person is dreaming.
121
386647
3582
06:44
Not just that ...
122
392333
1150
06:46
we can even get your breathing
while you are sitting like that,
123
394691
3087
06:49
and without touching you.
124
397802
1573
06:51
So he is sitting and reading
125
399399
1602
06:53
and this is his inhales, exhales.
126
401025
2945
06:55
We asked him to hold his breath,
127
403994
1991
06:58
and you see the signal
staying at a steady level
128
406009
2866
07:00
because he exhaled.
129
408899
1151
07:02
He did not inhale.
130
410074
1150
07:05
And I want to zoom in on the signal.
131
413279
2484
07:07
And this is the same signal as before.
132
415787
2063
07:09
These are the inhales,
133
417874
1681
07:11
these are the exhales.
134
419579
1644
07:13
And you see these blips on the signal?
135
421247
2087
07:15
These are not noise.
136
423358
1311
07:17
They are his heartbeats.
137
425159
1779
07:19
And you can see them beat by beat.
138
427971
1801
07:23
So I want to stop here for a moment
and show you a live demo.
139
431439
4221
07:27
Zach is going to help me with the demo,
140
435684
2071
07:29
and we're going to use the device
to monitor Zach's breathing.
141
437779
4755
07:35
So this white box
that you see here is the device,
142
443272
3768
07:39
and Zach is turning it on ...
143
447064
2341
07:43
and I see that he breathes well.
144
451581
2541
07:46
So we're going to do exactly what we did
in the video with the other guy,
145
454146
4602
07:50
so the wireless signal is going through,
146
458772
2691
07:53
it's touching Zach's body,
147
461487
1650
07:55
and it's reflecting back to the device,
148
463161
2223
07:57
and we want to monitor his breathing,
his inhale-exhale motion.
149
465408
3371
08:00
So we see the inhales, exhales --
150
468803
2523
08:03
so see, these ups and downs
are Zach breathing.
151
471350
4373
08:10
Inhaling, exhaling.
152
478191
2151
08:12
(Applause)
153
480366
5602
08:17
So, he can breathe.
154
485992
2152
08:20
(Laughter)
155
488168
1421
08:21
Zach, can you hold your breath, please?
156
489613
2198
08:25
OK, so now he's holding his breath,
157
493216
3063
08:28
so you see the signal stays
at a steady level,
158
496303
2254
08:30
and these are his heartbeats.
159
498581
1626
08:32
Beat, beat, beat, beat, beat.
160
500231
3476
08:35
(Applause)
161
503731
1001
08:36
OK, Zach, you can breathe again.
162
504756
2167
08:38
(Laughter)
163
506947
1889
08:40
We don't want accidents here.
164
508860
1420
08:42
(Laughter)
165
510304
1001
08:43
OK, thank you.
166
511329
1150
08:45
(Applause)
167
513756
5936
08:51
So as you can see, we have this device
168
519716
2761
08:54
that can monitor so many
physiological signals for you,
169
522501
3540
08:58
and what is really interesting
about this device
170
526065
2960
09:01
is that it does all this
without any wearables,
171
529049
2563
09:03
without asking the person
to change his behavior
172
531636
2698
09:06
or to wear anything
or charge anything special.
173
534358
2492
09:09
And that got doctors very excited,
174
537739
2968
09:12
because doctors,
175
540731
1151
09:13
they always want to know
more information about their patients,
176
541906
2957
09:16
particularly at home,
177
544887
1170
09:18
and this is particularly true
in chronic diseases,
178
546081
2634
09:20
like pulmonary diseases, like COPD,
179
548739
4155
09:24
or heart failure or Alzheimer's
and even depression.
180
552918
4524
09:29
All of these chronic diseases
are very important.
181
557466
2317
09:31
In fact -- perhaps you know --
182
559807
2468
09:34
two-thirds of the cost
of health care in the US
183
562299
2880
09:37
is due to chronic diseases.
184
565203
1984
09:39
But what is really interesting
about chronic diseases
185
567545
2531
09:42
is that when the person, for example,
186
570100
1820
09:43
has a problem that leads
to the hospital and the emergency room,
187
571944
4737
09:48
this problem doesn't happen overnight.
188
576705
2315
09:51
Actually, things happen gradually.
189
579044
2079
09:53
So if we can monitor
chronic disease patients in their home,
190
581147
3722
09:56
we can detect changes in their breathing,
heartbeat, mobility, sleep --
191
584893
4261
10:01
and we can detect emergencies
before they occur
192
589178
4182
10:05
and have the doctor intervene earlier
193
593384
2317
10:07
so that we can avoid hospitalization.
194
595725
2063
10:10
And indeed, today we are working
with multiple doctors
195
598645
4102
10:14
in different disease categories.
196
602771
1959
10:16
So I'm really excited
197
604754
1183
10:17
because we have deployed the device
with many patients.
198
605961
2684
10:20
We have deployed the device
with patients that have COPD,
199
608669
3332
10:24
which is a pulmonary disease,
200
612025
1507
10:25
patients that have Alzheimer's,
201
613556
2055
10:27
patients that have depression and anxiety
202
615635
2293
10:29
and people that have Parkinson's.
203
617952
1715
10:32
And we are working with the doctors
on improving their life,
204
620798
2883
10:35
understanding the disease better.
205
623705
1841
10:38
So when I started, I told you
206
626718
2571
10:41
that I'm really fascinated with Star Wars
and the Force in Star Wars,
207
629313
5420
10:46
and indeed, I'm still
very much fascinated,
208
634757
3326
10:50
even now, as a grown-up, with Star Wars,
209
638107
2006
10:52
waiting for the next movie.
210
640137
2295
10:54
But I'm very fascinated now and excited
211
642456
5269
10:59
about this new Force of wireless signals,
212
647749
3798
11:03
and the potential of changing
health care with this new force.
213
651571
4006
11:07
One of the patients with whom
we deployed is actually my aunt.
214
655601
2944
11:10
She has heart failure,
215
658569
2151
11:12
and I'm sure many of you guys
in the audience
216
660744
3708
11:16
have parents, grandparents,
loved ones who have chronic diseases.
217
664476
4688
11:21
So I want you to imagine with me a future
218
669188
2858
11:24
where in every home
that has a chronic disease patient,
219
672070
2882
11:26
there is a device like this device
sitting in the background
220
674976
3125
11:30
and just monitoring passively
221
678125
2365
11:32
sleep, breathing, the health
of this chronic disease patient,
222
680514
4666
11:37
and before an emergency occurs,
223
685204
2124
11:39
it would detect the degradation
in the physiological signal
224
687352
3023
11:42
and alert the doctor
225
690399
1500
11:43
so that we can avoid hospitalization.
226
691923
2063
11:46
This can change health care
as we know it today,
227
694652
3127
11:49
improve how we understand
chronic diseases
228
697803
3000
11:52
and also save many lives.
229
700827
1769
11:54
Thank you.
230
702898
1151
11:56
(Applause)
231
704073
5134
12:01
Helen Walters: Dina, thank you so much.
232
709231
1887
12:03
Thank you too, Zach.
233
711142
1151
12:04
So glad you're breathing.
234
712317
1311
12:05
So Dina, this is amazing.
235
713652
2697
12:08
The positive applications are incredible.
236
716373
3028
12:11
What is the framework, though,
like the ethical framework around this?
237
719425
3995
12:15
What are you doing to prevent
this technology from being used
238
723444
2880
12:18
for other, perhaps less positive
types of applications?
239
726348
3769
12:22
Dina Katabi: Yeah, this is
a very important question, of course,
240
730141
3055
12:25
like, what about misuse,
241
733220
1169
12:26
or what about, I guess you could say,
about the Dark Side of the Force?
242
734413
3622
12:30
HW: Right, right.
243
738059
1174
12:31
(Laughter)
244
739257
2087
12:33
DK: So we actually have technologies
245
741368
3499
12:36
that prevent people
from trying to use this device
246
744891
4277
12:41
to monitor somebody without their consent.
247
749192
2354
12:43
Because the device understands space,
248
751570
2253
12:45
it will ask you to prove,
by doing certain movements,
249
753847
3421
12:49
that you have access to the space
250
757292
1755
12:51
and you are the person
who you are asking the device to monitor.
251
759071
3300
12:54
So technology-wise,
252
762395
2190
12:56
we have technology
that we integrate to prevent misuse,
253
764609
3388
13:00
but also, I think there is a role
for policy, like everything else,
254
768021
3890
13:03
and hopefully, with the two of them,
we can control any misuse.
255
771935
5436
13:09
HW: Amazing. Thank you so much.
256
777395
1557
13:10
DK: Thank you.
257
778976
1151
13:12
(Applause)
258
780151
3769

▲Back to top

ABOUT THE SPEAKER
Dina Katabi - Technologist
Dina Katabi investigates how AI can make wireless devices sense human motion and vital signs.

Why you should listen

Dina Katabi designs new wireless devices that use machine learning to sense people through walls and occlusions. Her devices look like a Wi-Fi box. They transmit a low-power wireless signal and capture its reflections as it bounces off people and objects. They analyze those reflections to learn how people walk, measure their gait and detect elderly falls. The device can also measure a person's breathing, heart rate and sleep quality using wireless signals, without any sensor on the person's body. Katabi is working with medical doctors to use her technology to detect health emergencies and provide a better understanding of chronic diseases such as Alzheimer's and Parkinson's. 

Katabi is the Andrew & Erna Viterbi Professor of Electrical Engineering and Computer Science at MIT. She is also the director of the MIT's Center for Wireless Networks and Mobile Computing, a member of the National Academy of Engineering and a recipient of the MacArthur Fellowship. Her research has been recognized by the ACM Prize in Computing, the ACM Grace Murray Hopper Award, the SIGCOMM Test-of-Time Award, the IEEE William R. Bennett prize, the Faculty Research Innovation Fellowship, a Sloan Fellowship and multiple best paper awards. Several startups have been spun out of her lab, such as PiCharging and Emerald.

More profile about the speaker
Dina Katabi | Speaker | TED.com