TEDGlobal 2017
Pratik Shah: How AI is making it easier to diagnose disease
Pratik Shah: Como facilita a intelixencia artificial o diagnóstico de enfermidades
Filmed:
Readability: 6.9
1,571,835 views
Hoxe en día os algoritmos de intelixencia artificial (IA) necesitan decenas de miles de custosas imaxes médicas para detectar unha doenza nun paciente. E se puidésemos reducir de forma drástica a cantidade de datos requiridos para adestrar un sistema de IA, facendo así posibles diagnósticos de baixo custo e máis eficientes? Pratik Shah, membro TED, está a traballar nun enxeñoso sistema para facer exactamente iso. Cun enfoque de IA pouco convencional, Shah creou unha tecnoloxía que só necesita 50 imaxes para desenvolver un algoritmo que funcione --mesmo con fotografías tomadas polo propio médico co seu móbil-- para obter un diagnóstico. Descubre máis sobre como esta nova maneira de analizar a información médica podería facer posible a detección máis temperá de enfermidades potencialmente mortais e estender o diagnóstico asistido por IA a moitos lugares do mundo.
Pratik Shah - Medical technologist
Dr. Pratik Shah creates novel intersections between engineering, medical imaging, machine learning and medicine. Full bio
Dr. Pratik Shah creates novel intersections between engineering, medical imaging, machine learning and medicine. Full bio
Double-click the English transcript below to play the video.
00:13
Computer algorithms today
are performing incredible tasks
are performing incredible tasks
0
1280
3856
Hoxe en día, os algoritmos informáticos
realizan tarefas incribles
realizan tarefas incribles
00:17
with high accuracies, at a massive scale,
using human-like intelligence.
using human-like intelligence.
1
5160
4736
con gran precisión e a enorme escala,
amosando intelixencia similar á nosa.
amosando intelixencia similar á nosa.
00:21
And this intelligence of computers
is often referred to as AI
is often referred to as AI
2
9920
3936
A esta intelixencia informática
chámaselle a miúdo IA,
chámaselle a miúdo IA,
00:25
or artificial intelligence.
3
13880
1856
é dicir, intelixencia artificial.
00:27
AI is poised to make an incredible impact
on our lives in the future.
on our lives in the future.
4
15760
4200
A IA está lista para ter no futuro
un impacto incrible nas nosas vidas.
un impacto incrible nas nosas vidas.
00:32
Today, however,
we still face massive challenges
we still face massive challenges
5
20880
3936
Con todo, hoxe aínda temos que
enfrontarnos a enormes desafíos
enfrontarnos a enormes desafíos
00:36
in detecting and diagnosing
several life-threatening illnesses,
several life-threatening illnesses,
6
24840
3496
para detectar e diagnosticar varias
enfermidades potencialmente mortais,
enfermidades potencialmente mortais,
00:40
such as infectious diseases and cancer.
7
28360
2360
como as infecciosas ou o cancro.
00:44
Thousands of patients every year
8
32000
2296
Cada ano miles de pacientes
00:46
lose their lives
due to liver and oral cancer.
due to liver and oral cancer.
9
34320
2800
perden a vida a causa
do cancro de fígado ou de boca.
do cancro de fígado ou de boca.
00:49
Our best way to help these patients
10
37880
2696
O mellor modo de axudar a estes pacientes
00:52
is to perform early detection
and diagnoses of these diseases.
and diagnoses of these diseases.
11
40600
4320
é detectar e diagnosticar
a enfermidade en fases temperás.
a enfermidade en fases temperás.
00:57
So how do we detect these diseases today,
and can artificial intelligence help?
and can artificial intelligence help?
12
45880
4160
Como detectamos hoxe estas enfermidades?
Pode axudar a intelixencia artificial?
Pode axudar a intelixencia artificial?
01:03
In patients who, unfortunately,
are suspected of these diseases,
are suspected of these diseases,
13
51920
3656
Cando se sospeita que, por desgraza,
un paciente padece unha destas doenzas,
un paciente padece unha destas doenzas,
01:07
an expert physician first orders
14
55600
2656
un médico experto comeza por pedir
01:10
very expensive
medical imaging technologies
medical imaging technologies
15
58280
2616
probas carísimas baseadas
en tecnoloxías de imaxe,
en tecnoloxías de imaxe,
01:12
such as fluorescent imaging,
CTs, MRIs, to be performed.
CTs, MRIs, to be performed.
16
60920
4096
como a microscopia de fluorescencia,
a tomografía ou a resonancia magnética.
a tomografía ou a resonancia magnética.
01:17
Once those images are collected,
17
65040
2296
Unha vez obtidas esas imaxes,
01:19
another expert physician then diagnoses
those images and talks to the patient.
those images and talks to the patient.
18
67360
4520
outro experto fai un diagnóstico
e fala co paciente.
e fala co paciente.
01:24
As you can see, this is
a very resource-intensive process,
a very resource-intensive process,
19
72520
3456
Como vedes, é un proceso
que consome moitos recursos,
que consome moitos recursos,
01:28
requiring both expert physicians,
expensive medical imaging technologies,
expensive medical imaging technologies,
20
76000
4416
ao requirir médicos expertos
e custosas tecnoloxías médicas de imaxe,
e custosas tecnoloxías médicas de imaxe,
01:32
and is not considered practical
for the developing world.
for the developing world.
21
80440
3096
e non se considera práctico nos países
en vías de desenvolvemento.
en vías de desenvolvemento.
01:35
And in fact, in many
industrialized nations, as well.
industrialized nations, as well.
22
83560
3360
De feito, tampouco en moitas
nacións industrializadas.
nacións industrializadas.
01:39
So, can we solve this problem
using artificial intelligence?
using artificial intelligence?
23
87760
2880
Pódese resolver o problema
coa axuda da intelixencia artificial?
coa axuda da intelixencia artificial?
01:43
Today, if I were to use traditional
artificial intelligence architectures
artificial intelligence architectures
24
91840
4056
Se hoxe usásemos arquitecturas
de intelixencia artificial tradicionais
de intelixencia artificial tradicionais
para resolver este problema,
01:47
to solve this problem,
25
95920
1216
01:49
I would require 10,000 --
26
97160
1456
faría falla xerar 10.000,
01:50
I repeat, on an order of 10,000
of these very expensive medical images
of these very expensive medical images
27
98640
4016
repito, arredor de 10.000
destas imaxes médicas carísimas,
destas imaxes médicas carísimas,
01:54
first to be generated.
28
102680
1376
como primeiro paso.
01:56
After that, I would then go
to an expert physician,
to an expert physician,
29
104080
2896
Logo acudiría a un médico experto,
01:59
who would then analyze
those images for me.
those images for me.
30
107000
2496
que analizaría esas imaxes para min.
02:01
And using those two pieces of information,
31
109520
2096
E con esas dúas fontes de datos,
02:03
I can train a standard deep neural network
or a deep learning network
or a deep learning network
32
111640
3656
podo adestrar unha rede neural estándar
ou unha rede de aprendizaxe profunda
ou unha rede de aprendizaxe profunda
02:07
to provide patient's diagnosis.
33
115320
2136
para que faga o diagnóstico.
02:09
Similar to the first approach,
34
117480
1736
Ao igual que no primeiro método,
partir da intelixencia artificial
tradicional
tradicional
02:11
traditional artificial
intelligence approaches
intelligence approaches
35
119240
2143
02:13
suffer from the same problem.
36
121407
1449
presenta o mesmo problema.
02:14
Large amounts of data, expert physicians
and expert medical imaging technologies.
and expert medical imaging technologies.
37
122880
4560
Moitísimos datos, médicos expertos
e tecnoloxías de imaxe especializadas.
e tecnoloxías de imaxe especializadas.
02:20
So, can we invent more scalable, effective
38
128320
4296
É posible inventar arquitecturas
de intelixencia artificial
de intelixencia artificial
02:24
and more valuable artificial
intelligence architectures
intelligence architectures
39
132640
3296
ampliables, máis eficientes e máis útiles
02:27
to solve these very important
problems facing us today?
problems facing us today?
40
135960
3056
para resolver estes importantes problemas
que temos hoxe?
que temos hoxe?
Isto é precisamente o que fai o meu grupo
no Media Lab do MIT.
no Media Lab do MIT.
02:31
And this is exactly
what my group at MIT Media Lab does.
what my group at MIT Media Lab does.
41
139040
3296
02:34
We have invented a variety
of unorthodox AI architectures
of unorthodox AI architectures
42
142360
3856
Levamos inventadas varias arquitecturas
de IA pouco convencionais
de IA pouco convencionais
02:38
to solve some of the most important
challenges facing us today
challenges facing us today
43
146240
3176
para resolver algúns dos desafíos
actuais máis importantes
actuais máis importantes
02:41
in medical imaging and clinical trials.
44
149440
2200
no campo da imaxe médica
e as probas clínicas.
e as probas clínicas.
02:44
In the example I shared
with you today, we had two goals.
with you today, we had two goals.
45
152480
3056
No caso que veño de describirvos
tiñamos dous obxectivos.
tiñamos dous obxectivos.
02:47
Our first goal was to reduce
the number of images
the number of images
46
155560
2976
O primeiro era reducir o número de imaxes
02:50
required to train
artificial intelligence algorithms.
artificial intelligence algorithms.
47
158560
3256
que cómpren para adestrar
algoritmos de intelixencia artificial.
algoritmos de intelixencia artificial.
02:53
Our second goal -- we were more ambitious,
48
161840
2096
O segundo era máis ambicioso.
02:55
we wanted to reduce the use
of expensive medical imaging technologies
of expensive medical imaging technologies
49
163960
3736
Queríamos reducir o uso
de custosas tecnoloxías de imaxe médica
de custosas tecnoloxías de imaxe médica
na detección de enfermidades.
02:59
to screen patients.
50
167720
1216
03:00
So how did we do it?
51
168960
1200
Como o fixemos?
03:02
For our first goal,
52
170920
1216
Para o primeiro fin,
03:04
instead of starting
with tens and thousands
with tens and thousands
53
172160
2056
no canto de comezar con decenas de miles
03:06
of these very expensive medical images,
like traditional AI,
like traditional AI,
54
174240
3016
desas carísimas imaxes médicas,
como na IA tradicional,
como na IA tradicional,
03:09
we started with a single medical image.
55
177280
2056
empezamos con só unha.
03:11
From this image, my team and I
figured out a very clever way
figured out a very clever way
56
179360
3776
A partir desta imaxe, o meu equipo
atopou un modo moi enxeñoso
atopou un modo moi enxeñoso
03:15
to extract billions
of information packets.
of information packets.
57
183160
2736
de extraer miles de millóns
de paquetes de datos.
de paquetes de datos.
03:17
These information packets
included colors, pixels, geometry
included colors, pixels, geometry
58
185920
3696
Estes paquetes contiñan
cores, píxels, xeometría
cores, píxels, xeometría
03:21
and rendering of the disease
on the medical image.
on the medical image.
59
189640
2536
e representación da doenza
na imaxe médica.
na imaxe médica.
03:24
In a sense, we converted one image
into billions of training data points,
into billions of training data points,
60
192200
4336
En certo sentido, convertemos unha imaxe
en miles de millóns de observacións
en miles de millóns de observacións
03:28
massively reducing the amount of data
needed for training.
needed for training.
61
196560
3536
e reducimos moito a cantidade de datos
necesaria para adestrar o sistema.
necesaria para adestrar o sistema.
03:32
For our second goal,
62
200120
1216
Para o segundo obxectivo,
03:33
to reduce the use of expensive medical
imaging technologies to screen patients,
imaging technologies to screen patients,
63
201360
3856
reducir o uso de custosas tecnoloxías
de imaxe para detectar enfermidades,
de imaxe para detectar enfermidades,
03:37
we started with a standard,
white light photograph,
white light photograph,
64
205240
2856
comezamos cunha foto estándar
iluminada con luz branca,
iluminada con luz branca,
03:40
acquired either from a DSLR camera
or a mobile phone, for the patient.
or a mobile phone, for the patient.
65
208120
4336
tomada cunha cámara réflex dixital
ou cun móbil.
ou cun móbil.
Entón, lembrades os miles de millóns
de paquetes de datos?
de paquetes de datos?
03:44
Then remember those
billions of information packets?
billions of information packets?
66
212480
2456
03:46
We overlaid those from
the medical image onto this image,
the medical image onto this image,
67
214960
3536
Superpuxemos ese datos obtidos
da imaxe médica sobre estoutra imaxe,
da imaxe médica sobre estoutra imaxe,
03:50
creating something
that we call a composite image.
that we call a composite image.
68
218520
2520
creando así o que chamamos
unha imaxe composta.
unha imaxe composta.
03:53
Much to our surprise,
we only required 50 --
we only required 50 --
69
221480
3296
Para a nosa sorpresa, fixeron falla só 50,
03:56
I repeat, only 50 --
70
224800
1336
insisto, só 50,
03:58
of these composite images to train
our algorithms to high efficiencies.
our algorithms to high efficiencies.
71
226160
3840
desas imaxes compostas para adestrar
algoritmos ata taxas altas de eficiencia.
algoritmos ata taxas altas de eficiencia.
04:02
To summarize our approach,
72
230680
1336
Para resumir o noso enfoque,
04:04
instead of using 10,000
very expensive medical images,
very expensive medical images,
73
232040
3176
no canto de empregar 10.000
imaxes médicas carísimas,
imaxes médicas carísimas,
04:07
we can now train the AI algorithms
in an unorthodox way,
in an unorthodox way,
74
235240
3016
agora podemos adestrar os algoritmos
de modo pouco convencional,
de modo pouco convencional,
04:10
using only 50 of these high-resolution,
but standard photographs,
but standard photographs,
75
238280
4256
usando só 50 destas fotografías,
de alta resolución pero estándares,
de alta resolución pero estándares,
04:14
acquired from DSLR cameras
and mobile phones,
and mobile phones,
76
242560
2496
tomadas con réflex dixitais
ou teléfonos móbiles,
ou teléfonos móbiles,
04:17
and provide diagnosis.
77
245080
1536
e obter un diagnóstico.
04:18
More importantly,
78
246640
1216
E o que é máis importante,
04:19
our algorithms can accept,
in the future and even right now,
in the future and even right now,
79
247880
3136
os nosos algoritmos admiten,
no futuro e xa agora mesmo,
no futuro e xa agora mesmo,
04:23
some very simple, white light
photographs from the patient,
photographs from the patient,
80
251040
2816
fotografías sinxelas, de luz branca,
feitas polo paciente,
feitas polo paciente,
04:25
instead of expensive
medical imaging technologies.
medical imaging technologies.
81
253880
2440
no canto de tecnoloxías de imaxe médica
moi custosas.
moi custosas.
04:29
I believe that we are poised
to enter an era
to enter an era
82
257120
3096
Penso que estamos xa listos
para entrar nunha era
para entrar nunha era
04:32
where artificial intelligence
83
260240
1936
na que a intelixencia artificial
04:34
is going to make an incredible
impact on our future.
impact on our future.
84
262200
2536
vai ter un impacto incrible
no noso futuro.
no noso futuro.
04:36
And I think that thinking
about traditional AI,
about traditional AI,
85
264760
2456
E en relación coa IA tradicional,
04:39
which is data-rich but application-poor,
86
267240
2776
potente no manexo de datos pero
moi débil nas aplicacións,
moi débil nas aplicacións,
04:42
we should also continue thinking
87
270040
1536
deberíamos seguir pensando tamén
04:43
about unorthodox artificial
intelligence architectures,
intelligence architectures,
88
271600
3016
en arquitecturas de intelixencia
artificial pouco convencionais
artificial pouco convencionais
04:46
which can accept small amounts of data
89
274640
1936
que poden traballar con poucos datos
para resolver algúns
dos maiores problemas que temos hoxe,
dos maiores problemas que temos hoxe,
04:48
and solve some of the most important
problems facing us today,
problems facing us today,
90
276600
2936
04:51
especially in health care.
91
279560
1256
sobre todo en sanidade.
04:52
Thank you very much.
92
280840
1216
Moitas grazas.
04:54
(Applause)
93
282080
3840
(Aplausos)
ABOUT THE SPEAKER
Pratik Shah - Medical technologistDr. Pratik Shah creates novel intersections between engineering, medical imaging, machine learning and medicine.
Why you should listen
Dr. Shah's research program at the MIT Media Lab develops scalable and low-cost diagnostics and therapeutics. His ongoing research areas at MIT include: 1) artificial intelligence and machine learning methods for detection of cancer biomarkers using standard photographs vs. expensive medical images; 2) unorthodox artificial intelligence and machine learning algorithms to design optimal and faster clinical trials and to reduce adverse effects on patients; and 3) low-cost and open source imaging devices, paper diagnostics, algorithms and mobile phones to improve public health and generate real-world data.
Clinical studies with Pratik's medical technologies have revealed "missing sick" patients, who otherwise remain undiagnosed in conventional healthcare settings. Dr. Shah's graduate and postdoctoral research contributed to the discovery of a vaccine component to prevent pneumococcal (Streptococcus pneumoniae) diseases; the identification of new pathways, technologies and metabolites as antimicrobials to target gastrointestinal infections; and a nonprofit to deploy a low-cost water quality test for the developing world.
Past recognition for Dr. Shah includes the American Society for Microbiology's Raymond W. Sarber national award, the Harvard Medical School and Massachusetts General Hospitals ECOR Fund for Medical Discovery postdoctoral fellowship, the AAAS-Lemelson Invention Ambassador Award and a TED Fellowship. Pratik has been an invited discussion leader at Gordon Research Seminars; a speaker at Cold Spring Harbor Laboratories, Gordon Research Conferences and IEEE bioengineering conferences; and a peer reviewer for leading scientific publications and funding agencies. Pratik has a BS, MS, and a PhD in Microbiology and completed fellowship training at The Broad Institute of MIT and Harvard, Massachusetts General Hospital and Harvard Medical School.
Pratik Shah | Speaker | TED.com