ABOUT THE SPEAKER
Pratik Shah - Medical technologist
Dr. 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.

More profile about the speaker
Pratik Shah | Speaker | TED.com
TEDGlobal 2017

Pratik Shah: How AI is making it easier to diagnose disease

Pratik Shah: 人工智能:让诊断疾病变得更容易

Filmed:
1,571,835 views

今天的AI算法,需要成千上万昂贵的医学图像来检测患者的疾病。 怎样才能大幅减少训练AI所需的数据量,使诊断成本更低且更有效? TED研究员Pratik Shah正在研究一个聪明的系统来做到这一点。 使用非正统的人工智能方法,Shah开发了一种技术,只需要50张图像就可以开发出一种有效算法——甚至可以使用在医生手机上拍摄的照片来提供诊断。 详细了解这种分析医疗信息的新方法如何更早发现危及生命的疾病,并将AI辅助诊断带到全球更多的医疗机构。
- Medical technologist
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电脑 algorithms算法 today今天
are performing执行 incredible难以置信 tasks任务
0
1280
3856
今天的计算机算法,
正在使用类似人类的智能,
大规模的执行具有高精度的,
不可思议的任务。
00:17
with high accuracies精度, at a massive大规模的 scale规模,
using运用 human-like类人 intelligence情报.
1
5160
4736
00:21
And this intelligence情报 of computers电脑
is often经常 referred简称 to as AIAI
2
9920
3936
而这种计算机智能,通常被称为AI,
00:25
or artificial人造 intelligence情报.
3
13880
1856
或“人工智能”。
00:27
AIAI is poised准备 to make an incredible难以置信 impact碰撞
on our lives生活 in the future未来.
4
15760
4200
人工智能有望在未来对我们的生活
产生令人难以置信的影响。
00:32
Today今天, however然而,
we still face面对 massive大规模的 challenges挑战
5
20880
3936
然而今天,在检测和诊断
几种危及生命的疾病,
00:36
in detecting检测 and diagnosing诊断
several一些 life-threatening危及生命 illnesses疾病,
6
24840
3496
比如传染病和癌症时,
00:40
such这样 as infectious传染病 diseases疾病 and cancer癌症.
7
28360
2360
我们仍然面临着大量的挑战。
00:44
Thousands成千上万 of patients耐心 every一切 year
8
32000
2296
每年,数以千计的病人
00:46
lose失去 their lives生活
due应有 to liver and oral口服 cancer癌症.
9
34320
2800
因患上肝癌和口腔癌失去生命。
00:49
Our best最好 way to help these patients耐心
10
37880
2696
帮助病人最好的方式
就是对这些疾病进行
早期检测和诊断。
00:52
is to perform演出 early detection发现
and diagnoses诊断 of these diseases疾病.
11
40600
4320
00:57
So how do we detect检测 these diseases疾病 today今天,
and can artificial人造 intelligence情报 help?
12
45880
4160
那么,今天我们如何检测这些疾病?
AI可以提供帮助吗?
01:03
In patients耐心 who, unfortunately不幸,
are suspected嫌疑 of these diseases疾病,
13
51920
3656
对于不幸被怀疑
患有这些疾病的患者,
专家医师会先要求他们照射
01:07
an expert专家 physician医师 first orders命令
14
55600
2656
01:10
very expensive昂贵
medical imaging成像 technologies技术
15
58280
2616
非常昂贵的医疗图像,
例如荧光成像,CT,MRI等。
01:12
such这样 as fluorescent imaging成像,
CTsCts, MRIs核磁共振成像, to be performed执行.
16
60920
4096
收集到这些图像之后,
01:17
Once一旦 those images图片 are collected,
17
65040
2296
另一位专家医师会进行诊断,
并与患者交流。
01:19
another另一个 expert专家 physician医师 then diagnoses诊断
those images图片 and talks会谈 to the patient患者.
18
67360
4520
01:24
As you can see, this is
a very resource-intensive资源密集型 process处理,
19
72520
3456
显而易见,这是个
非常耗费资源的过程,
需要两位专家医师
和昂贵的医学图像技术。
01:28
requiring要求 both expert专家 physicians医师,
expensive昂贵 medical imaging成像 technologies技术,
20
76000
4416
01:32
and is not considered考虑 practical实际的
for the developing发展 world世界.
21
80440
3096
这在发展中国家被认为并不实用,
事实上,在许多
工业化国家也是如此。
01:35
And in fact事实, in many许多
industrialized工业化 nations国家, as well.
22
83560
3360
01:39
So, can we solve解决 this problem问题
using运用 artificial人造 intelligence情报?
23
87760
2880
那么,我们能够用
人工智能解决这个问题吗?
01:43
Today今天, if I were to use traditional传统
artificial人造 intelligence情报 architectures架构
24
91840
4056
今天,如果使用传统的
人工智能架构
01:47
to solve解决 this problem问题,
25
95920
1216
来解决这个问题,
我可能需要1万张——
01:49
I would require要求 10,000 --
26
97160
1456
我重复一次,我首先需要
生成1万张这种非常昂贵的
01:50
I repeat重复, on an order订购 of 10,000
of these very expensive昂贵 medical images图片
27
98640
4016
医学图像。
01:54
first to be generated产生.
28
102680
1376
之后,我会去找一位专业医师
01:56
After that, I would then go
to an expert专家 physician医师,
29
104080
2896
01:59
who would then analyze分析
those images图片 for me.
30
107000
2496
为我分析这些图像。
利用这两条信息,
02:01
And using运用 those two pieces of information信息,
31
109520
2096
02:03
I can train培养 a standard标准 deep neural神经 network网络
or a deep learning学习 network网络
32
111640
3656
我可以训练标准的深度神经网络,
或深度学习网络
02:07
to provide提供 patient's耐心 diagnosis诊断.
33
115320
2136
对患者进行诊断。
02:09
Similar类似 to the first approach途径,
34
117480
1736
与第一步相似,
02:11
traditional传统 artificial人造
intelligence情报 approaches方法
35
119240
2143
传统人工智能方法
02:13
suffer遭受 from the same相同 problem问题.
36
121407
1449
遭遇了同样的问题:
02:14
Large amounts of data数据, expert专家 physicians医师
and expert专家 medical imaging成像 technologies技术.
37
122880
4560
那就是需要大量的数据、
专家医师和专业的医疗图像技术。
02:20
So, can we invent发明 more scalable可扩展性, effective有效
38
128320
4296
我们是否能够创造出一种
规模更大、更有效率、
同时更有价值的人工智能架构,
02:24
and more valuable有价值 artificial人造
intelligence情报 architectures架构
39
132640
3296
02:27
to solve解决 these very important重要
problems问题 facing面对 us today今天?
40
135960
3056
来解决我们今天面临的
这些重要的问题呢?
而这就是我们的团队
在MIT媒体实验室所研究的内容。
02:31
And this is exactly究竟
what my group at MITMIT Media媒体 Lab实验室 does.
41
139040
3296
我们开发了各种新型AI架构,
02:34
We have invented发明 a variety品种
of unorthodox非正统的 AIAI architectures架构
42
142360
3856
来解决一些我们当今
在医疗图像和临床试验中
02:38
to solve解决 some of the most important重要
challenges挑战 facing面对 us today今天
43
146240
3176
02:41
in medical imaging成像 and clinical临床 trials试验.
44
149440
2200
面临的最重要的挑战。
02:44
In the example I shared共享
with you today今天, we had two goals目标.
45
152480
3056
在我今天分享的例子中,
包括了我们的两个目标。
第一个目标,是减少
用来训练人工智能算法
02:47
Our first goal目标 was to reduce减少
the number of images图片
46
155560
2976
所需要的图片数量。
02:50
required需要 to train培养
artificial人造 intelligence情报 algorithms算法.
47
158560
3256
第二个目标——更大的志向,
02:53
Our second第二 goal目标 -- we were more ambitious有雄心,
48
161840
2096
02:55
we wanted to reduce减少 the use
of expensive昂贵 medical imaging成像 technologies技术
49
163960
3736
我们希望让患者减少使用昂贵的
医疗图像技术。
02:59
to screen屏幕 patients耐心.
50
167720
1216
那么我们是怎样做的?
03:00
So how did we do it?
51
168960
1200
03:02
For our first goal目标,
52
170920
1216
我们的第一个目标,
03:04
instead代替 of starting开始
with tens and thousands数千
53
172160
2056
相比于传统AI
从成千上万张昂贵的医疗图像开始,
03:06
of these very expensive昂贵 medical images图片,
like traditional传统 AIAI,
54
174240
3016
我们选择从单张图像开始。
03:09
we started开始 with a single medical image图片.
55
177280
2056
根据这张图片,
我和我的团队想出了
03:11
From this image图片, my team球队 and I
figured想通 out a very clever聪明 way
56
179360
3776
一种非常聪明的方法
来提取数十亿个信息包。
03:15
to extract提取 billions数十亿
of information信息 packets.
57
183160
2736
这些信息包包含颜色、像素、形态
03:17
These information信息 packets
included包括 colors颜色, pixels像素, geometry几何
58
185920
3696
03:21
and rendering翻译 of the disease疾病
on the medical image图片.
59
189640
2536
和疾病呈现在医疗图像上的效果。
这样一来,我们就将一张图像
转换成了数十亿个训练数据点,
03:24
In a sense, we converted转换 one image图片
into billions数十亿 of training训练 data数据 points,
60
192200
4336
03:28
massively大规模 reducing减少 the amount of data数据
needed需要 for training训练.
61
196560
3536
需要训练的数据量就大大减少了。
03:32
For our second第二 goal目标,
62
200120
1216
第二个目标,
是减少对患者使用医疗图像技术。
03:33
to reduce减少 the use of expensive昂贵 medical
imaging成像 technologies技术 to screen屏幕 patients耐心,
63
201360
3856
最开始,我们会从
数码单反相机或手机中
03:37
we started开始 with a standard标准,
white白色 light photograph照片,
64
205240
2856
获取一张标准的白色光线照片。
03:40
acquired后天 either from a DSLRDSLR camera相机
or a mobile移动 phone电话, for the patient患者.
65
208120
4336
然后,还记得那
数十亿个信息包吗?
03:44
Then remember记得 those
billions数十亿 of information信息 packets?
66
212480
2456
将这些医疗图像的信息包
覆盖在这张图片上,
03:46
We overlaid覆盖 those from
the medical image图片 onto this image图片,
67
214960
3536
这时我们就创建了一张合成图像。
03:50
creating创建 something
that we call a composite综合 image图片.
68
218520
2520
03:53
Much to our surprise,
we only required需要 50 --
69
221480
3296
令人惊讶的是,我们只需要50张——
强调一下,仅仅50张——
03:56
I repeat重复, only 50 --
70
224800
1336
这些复合图像,
就能训练我们的算法提高效率。
03:58
of these composite综合 images图片 to train培养
our algorithms算法 to high efficiencies效率.
71
226160
3840
04:02
To summarize总结 our approach途径,
72
230680
1336
总结一下我们的方法,
04:04
instead代替 of using运用 10,000
very expensive昂贵 medical images图片,
73
232040
3176
区别于用1万张昂贵的
医疗图像训练AI算法,
我们使用了一种全新的方式,
04:07
we can now train培养 the AIAI algorithms算法
in an unorthodox非正统的 way,
74
235240
3016
只需要将数码相机或手机拍摄的
04:10
using运用 only 50 of these high-resolution高分辨率,
but standard标准 photographs照片,
75
238280
4256
50张高分辨率的标准照片,
04:14
acquired后天 from DSLRDSLR cameras相机
and mobile移动 phones手机,
76
242560
2496
04:17
and provide提供 diagnosis诊断.
77
245080
1536
即可提供诊断。
04:18
More importantly重要的,
78
246640
1216
更重要的是,
在未来,甚至现在,
我们的算法可以接受
04:19
our algorithms算法 can accept接受,
in the future未来 and even right now,
79
247880
3136
一些病人自己拍摄的白光照片,
04:23
some very simple简单, white白色 light
photographs照片 from the patient患者,
80
251040
2816
来替代昂贵的医疗图像技术。
04:25
instead代替 of expensive昂贵
medical imaging成像 technologies技术.
81
253880
2440
04:29
I believe that we are poised准备
to enter输入 an era时代
82
257120
3096
我相信,我们已经准备好
进入这样一个时代,
人工智能
04:32
where artificial人造 intelligence情报
83
260240
1936
正在对我们的未来产生
不可思议的影响。
04:34
is going to make an incredible难以置信
impact碰撞 on our future未来.
84
262200
2536
我也认为相比拥有丰富数据
04:36
And I think that thinking思维
about traditional传统 AIAI,
85
264760
2456
但应用困难的传统AI,
04:39
which哪一个 is data-rich数据丰富 but application-poor应用程序穷,
86
267240
2776
04:42
we should also continue继续 thinking思维
87
270040
1536
我们应该不断思考
非传统的人工智能架构。
04:43
about unorthodox非正统的 artificial人造
intelligence情报 architectures架构,
88
271600
3016
它们能够接受少量数据,
04:46
which哪一个 can accept接受 small amounts of data数据
89
274640
1936
并解决一些今天
我们所面临的重要问题,
04:48
and solve解决 some of the most important重要
problems问题 facing面对 us today今天,
90
276600
2936
特别是在医疗健康方面。
04:51
especially特别 in health健康 care关心.
91
279560
1256
非常感谢。
04:52
Thank you very much.
92
280840
1216
(掌声)
04:54
(Applause掌声)
93
282080
3840
Translated by Shuhui Chen

▲Back to top

ABOUT THE SPEAKER
Pratik Shah - Medical technologist
Dr. 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.

More profile about the speaker
Pratik Shah | Speaker | TED.com