How AI is Revolutionizing Medicine — Transcript

Full transcript

1. 0:00When we talk about AI and healthcare,
2. 0:03I think we need to understand it as a foundational change in the toolkit we have
3. 0:07available In the same way, and I mean this profoundly,
4. 0:11that algebra changed our understanding of the world and of math.
5. 0:15In our field. The major changes are happening in diagnostic testing,
6. 0:19how we can now interpret them more effectively and without as much expertise
7. 0:23needed.
8. 0:24We are seeing the old patterns of care change in the new ways of doing care.
9. 0:29The topic of how is AI going to be used is so interesting because these
10. 0:34new machine learning sophisticated models, they aren't the end of the question.
11. 0:39It's the beginning of the question.
12. 0:50AI is the future of healthcare. Or so we've been told. But what does it mean?
13. 0:55Is it coming for your job? Is it saving your life?
14. 0:58What risks does it pose and how does it have the potential to affect
15. 1:03you?
16. 1:04Artificial intelligence has in some form been around for decades,
17. 1:07but most people are just finding out about it because of things like ChatGPT,
18. 1:11which have come into the mainstream.
19. 1:14In healthcare artificial intelligence has been around for years
20. 1:17as well. It's helped doctors transcribe notes faster.
21. 1:21It's helped with patient experiences with chatbots, but with generative AI,
22. 1:25it's really supercharging what healthcare professionals and scientists are able
23. 1:28to do by helping predict future health risks,
24. 1:32make diagnoses faster and more accurately,
25. 1:35and also helping with potential drug discovery for new novel treatments and
26. 1:40therapeutics.
27. 1:41The AI boom really hit healthcare in the last five years.
28. 1:45We've actually been able to transition into having the models or the tools we
29. 1:50built to actually work with the data very effectively and have been able to
30. 1:54define new ways to treat diseases, new ways to diagnose diseases,
31. 1:58and finally find individualized signatures of how I as a patient or somebody
32. 2:03else as a patient would do if they got a certain treatment or therapy.
33. 2:07The Gen AI boom is assisting doctors in more advanced ways,
34. 2:11but both clinicians and patients are questioning the risks these new
35. 2:14technologies may bring with them.
36. 2:17I think a lot of people are wondering,
37. 2:19what's this going to look like when I go to the hospital,
38. 2:21when I go to the doctor? How's it going to be used to inform decisions?
39. 2:25What about errors, false positive, false negatives?
40. 2:28Who's going to be responsible?
41. 2:30There's a lot of questions around this that really begs the expertise of people
42. 2:35in different areas.
43. 2:36So we have agencies like the US Food and Drug Administration that does oversee
44. 2:41certain uses of AI in healthcare.
45. 2:43There are also groups like the American Medical Association and the National
46. 2:47Academy of Medicine that do put out guidelines and codes of conduct for
47. 2:51responsible use of AI.
48. 2:53So right now there are experts and there are various groups that are trying to
49. 2:56oversee this and make sure that it's used responsibly.
50. 2:59So there's this big tension about how AI will affect our workflows.
51. 3:04Clinicians have the ability to interact with patients,
52. 3:07understand their preferences,
53. 3:09so I still feel they'll have a key role to play in interpreting the outputs that
54. 3:13the AI technology gives them.
55. 3:17While the larger AI labs from Google to OpenAI are trying to disrupt the
56. 3:21healthcare industry at large,
57. 3:23there are smaller and more focused labs within academia dedicated to applying
58. 3:28AI to specialties.
59. 3:30Well, we assessed how well the model performed looking at different views.
60. 3:34We're seeing an increasing investment in AI and healthcare. Since 2020,
61. 3:38there has been a substantial increase in venture capital funding coming into
62. 3:42this space. In 2024,
63. 3:45we saw around $11 billion in funding going to various different AI and
64. 3:49healthcare startups.
65. 3:51Experts say that academics have different priorities than some of these AI
66. 3:55startups. Academics can fine tune large language models,
67. 3:58and they may have a different incentive to develop ethics guidelines around
68. 4:02their responsible use.
69. 4:04I think our biggest risk with AI is making sure we use it responsibly and
70. 4:08consistently.
71. 4:09We want people to pursue this creativity to the ends of the earth,
72. 4:15but we also need clear processes in place to ensure that before they ever make
73. 4:19it in front of a patient,
74. 4:21they have been through a series of stage gates and reviews to ensure they are
75. 4:25safe, effective, fair and equitable.
76. 4:30Yale has invested $150 million towards AI development
77. 4:35over the next five years.
78. 4:37One of the ongoing efforts at Yale Medicine is the cardiovascular data science
79. 4:41lab founded by Dr. Rohan carer in 2020.
80. 4:45So this is the cardiovascular data science lab or the cards lab.
81. 4:48It's not a single domain of AI that we work on.
82. 4:50We work on diagnostics on precision care or defining therapies that work for
83. 4:55individuals, work for cardiac imaging,
84. 4:58and other ways that we use to enable better care for patients.
85. 5:02A good example is electrocardiograms,
86. 5:04which are electrical recordings of the heart taken with stickers on the chest
87. 5:07available world over. ECGGPT,
88. 5:10which is a play on chatGPT.
89. 5:13It takes an ECG image and generates a report that is an example of generative
90. 5:17ai. We can generate a full report directly from the electrocardiogram.
91. 5:20Originally, this would require a cardiologist to interpret these studies,
92. 5:24but our hope is that this can form the basis of very accurate reads that are
93. 5:29available all over that eventually clinicians can confirm and use in their care.
94. 5:32So we decided to focus on app based solutions rather than things that integrate
95. 5:37in large health systems so that these tools are truly accessible. Here we have.
96. 5:42A typical 12 BDCG,
97. 5:44and we know me as a human reader I can see these are the leads coming in from
98. 5:48the chest. But for me, looking at this,
99. 5:50I could never tell you this is a patient with heart failure just based on the
100. 5:53ECG. What the AI eyeball is able to do is to kind of isolate these two leads and
101. 5:57detect some signal that triggers that yep,
102. 5:59this is someone with left ventricular systolic dysfunction,
103. 6:01an ejection fraction of less than 40%.
104. 6:04And it visualizes it in a way that a human reader wouldn't be able to detect.
105. 6:08While facilities like the Cards lab are developing these new AI tools at a rapid
106. 6:12pace,
107. 6:13there are still many steps needed before these technologies can be implemented
108. 6:17to ensure the public wellbeing.
109. 6:19Regulators may wish to impose oversight and guardrails before AI becomes
110. 6:24the new standard practice of care.
111. 6:27In terms of AI and misdiagnosis,
112. 6:30experts say that the cases of an examples of this are pretty limited at this
113. 6:34point because even though an AI might make a mistake or might make a
114. 6:38misdiagnosis,
115. 6:39there is still a real doctor involved in the diagnosis or prescription of a
116. 6:43medication.
117. 6:44So are we really going to become healthier?
118. 6:46I think it's not just the million dollar question,
119. 6:49it's the million life question. There is benefit to technology,
120. 6:53there is benefit to AI systems,
121. 6:56but it really requires that all of us,
122. 6:58we need to be able to access. We need to be able to use this.
123. 7:03We need to be willing to use it,
124. 7:05and we need to have that space or energy to do it.
125. 7:10On the wall of my office, I have four photographs. It's the transistor,
126. 7:14the camera, the telephone, and the source code for the worldwide web,
127. 7:18because to me,
128. 7:18those are the enabling innovations in my lifetime that transform the way we
129. 7:23deliver care - telemedicine, electronic health records,
130. 7:28MRI imaging, all of that derives from those foundational innovations.
131. 7:33I think AI is in that category.