1 00:00:00,719 --> 00:00:04,000 Exciting things are happening at Becker's Healthcare. Stay 2 00:00:04,000 --> 00:00:06,240 ahead of industry trends with the new Becker's 3 00:00:06,240 --> 00:00:08,419 CFO plus Revenue Cycle podcast, 4 00:00:08,800 --> 00:00:11,119 your go to source for insights from top 5 00:00:11,119 --> 00:00:14,154 healthcare finance leaders. Tune in wherever you get 6 00:00:14,154 --> 00:00:15,054 your podcasts. 7 00:00:15,914 --> 00:00:18,475 And don't miss the tenth annual health IT 8 00:00:18,475 --> 00:00:22,154 plus digital health plus RCM conference, happening September 9 00:00:22,154 --> 00:00:24,474 30 to 10/03/2025 10 00:00:24,474 --> 00:00:25,214 in Chicago. 11 00:00:25,755 --> 00:00:26,879 Join thousands of 12 00:00:27,279 --> 00:00:30,079 executives, engage with industry leaders, and explore the 13 00:00:30,079 --> 00:00:31,619 future of health care innovation. 14 00:00:32,320 --> 00:00:36,719 Learn more about our upcoming events at beckershospitalreview.com. 15 00:00:36,719 --> 00:00:37,539 See you there. 16 00:00:38,159 --> 00:00:40,320 This is Laura Dirda with the Becker's Healthcare 17 00:00:40,320 --> 00:00:40,820 podcast. 18 00:00:41,204 --> 00:00:43,125 I'm thrilled today to be joined by doctor 19 00:00:43,125 --> 00:00:46,245 Alvin Liu, endowed professor in AI and AI 20 00:00:46,245 --> 00:00:48,585 oversight team member as well as the inaugural 21 00:00:48,804 --> 00:00:49,304 director 22 00:00:49,685 --> 00:00:52,484 at the James p Gills junior MD and 23 00:00:52,484 --> 00:00:55,445 Heather Gills artificial intelligence innovation center at Johns 24 00:00:55,445 --> 00:00:56,344 Hopkins Medicine. 25 00:00:56,770 --> 00:00:58,049 Doctor Lewis, a pleasure to have you on 26 00:00:58,049 --> 00:00:59,030 the podcast today. 27 00:00:59,409 --> 00:01:01,570 Good morning, Laura. Thanks for having me. It's 28 00:01:01,570 --> 00:01:02,149 a pleasure. 29 00:01:02,530 --> 00:01:04,770 Absolutely. Now I'm really looking forward to this 30 00:01:04,770 --> 00:01:07,010 conversation because I know the work you're doing 31 00:01:07,010 --> 00:01:10,049 in AI and especially at Johns Hopkins really 32 00:01:10,049 --> 00:01:10,549 truly, 33 00:01:11,194 --> 00:01:13,194 is on the cutting edge of innovation and 34 00:01:13,194 --> 00:01:16,174 health care technology today. And so many organizations 35 00:01:16,395 --> 00:01:18,715 are trying to figure out how they can, 36 00:01:19,115 --> 00:01:21,515 use AI most effectively and what it means 37 00:01:21,515 --> 00:01:23,215 for their health systems and hospitals. 38 00:01:23,759 --> 00:01:25,859 So this will be a really fun conversation. 39 00:01:26,000 --> 00:01:28,239 But before we dig in, I'm wondering, first, 40 00:01:28,239 --> 00:01:30,399 could you introduce yourself, tell us a little 41 00:01:30,399 --> 00:01:31,920 bit about your background, and, 42 00:01:32,319 --> 00:01:35,379 what makes your role at, Johns Hopkins unique? 43 00:01:36,335 --> 00:01:38,895 Sure. Happy to do so. So, again, my 44 00:01:38,895 --> 00:01:40,174 name is Alvin Liu. 45 00:01:40,575 --> 00:01:42,734 I did most of my medical training at 46 00:01:42,734 --> 00:01:44,034 Johns Hopkins University. 47 00:01:44,734 --> 00:01:47,534 I'm a practicing retinal surgeon. And after my 48 00:01:47,534 --> 00:01:48,034 training, 49 00:01:48,599 --> 00:01:51,000 I stayed on as the medical faculty at 50 00:01:51,000 --> 00:01:51,819 Johns Hopkins. 51 00:01:52,599 --> 00:01:55,799 Currently, outside of my clinical work, I wear 52 00:01:55,799 --> 00:01:58,939 several hats which are all related to artificial 53 00:01:59,000 --> 00:01:59,500 intelligence. 54 00:02:00,359 --> 00:02:00,859 First, 55 00:02:01,314 --> 00:02:02,854 as a clinician scientist, 56 00:02:03,314 --> 00:02:06,135 I'm the inaugural director of the Wilmer Gilkes 57 00:02:06,275 --> 00:02:07,655 AI Innovation Center, 58 00:02:08,034 --> 00:02:10,694 which is the first endowed AI center 59 00:02:10,995 --> 00:02:13,814 at the Johns Hopkins University School of Medicine, 60 00:02:14,354 --> 00:02:16,935 which was made possible a few months ago 61 00:02:17,210 --> 00:02:20,349 by a very generous donation of $10,000,000. 62 00:02:21,210 --> 00:02:23,870 Second, I have been involved at Johns Hopkins 63 00:02:24,330 --> 00:02:27,290 in the deployments of AI tools for both 64 00:02:27,290 --> 00:02:29,310 clinical and operational processes, 65 00:02:29,844 --> 00:02:30,664 such as autonomous 66 00:02:31,125 --> 00:02:33,544 AI for diabetic retinopathy screening 67 00:02:34,084 --> 00:02:35,944 and revenue cycle management. 68 00:02:36,485 --> 00:02:39,784 And third, I serve on the AI oversight 69 00:02:39,844 --> 00:02:40,344 team, 70 00:02:40,724 --> 00:02:44,280 which is a system wide leadership team with 71 00:02:44,280 --> 00:02:46,620 purview over all things AI related 72 00:02:47,080 --> 00:02:49,979 in both the clinical and operational domains 73 00:02:50,280 --> 00:02:53,580 across the entire Johns Hopkins Medicine health system. 74 00:02:54,520 --> 00:02:56,439 That's great to hear. And, you know, really, 75 00:02:56,680 --> 00:02:58,935 cool projects that you're working on as well. 76 00:02:59,574 --> 00:03:00,854 I'd love to dig a little bit deeper 77 00:03:00,854 --> 00:03:02,935 in there. What are some of the things 78 00:03:02,935 --> 00:03:04,854 you're doing right now, the opportunities as well 79 00:03:04,854 --> 00:03:06,614 as the headwinds that you had your eye 80 00:03:06,614 --> 00:03:07,114 on? 81 00:03:07,974 --> 00:03:09,514 I would say across 82 00:03:10,055 --> 00:03:12,215 the different trends that I see in health 83 00:03:12,215 --> 00:03:14,395 care right now, artificial intelligence 84 00:03:15,069 --> 00:03:17,090 is currently the biggest opportunity 85 00:03:17,790 --> 00:03:20,930 and will remain the leading force for change 86 00:03:20,990 --> 00:03:22,449 in the foreseeable future. 87 00:03:22,909 --> 00:03:25,889 However, despite the generally rosy sentiments, 88 00:03:26,830 --> 00:03:30,055 significant headwinds do exist for real world scaling 89 00:03:30,055 --> 00:03:31,835 of these clinical AI tools. 90 00:03:32,135 --> 00:03:35,495 If you look at, a recently published report 91 00:03:35,495 --> 00:03:36,314 by the FDA, 92 00:03:36,855 --> 00:03:39,115 there are now over 1,000 93 00:03:39,735 --> 00:03:42,775 AI enabled medical devices that have been approved 94 00:03:42,775 --> 00:03:43,594 by the FDA. 95 00:03:44,250 --> 00:03:46,750 Yet if you look at real world deployment 96 00:03:46,810 --> 00:03:47,629 of these tools, 97 00:03:48,010 --> 00:03:50,349 we are still in the infancy stage. 98 00:03:50,730 --> 00:03:52,349 I think the key uncertainty 99 00:03:52,730 --> 00:03:55,530 is who and how are we go going 100 00:03:55,530 --> 00:03:57,550 to pay for these clinical AI tools. 101 00:03:58,435 --> 00:04:00,534 From, let's say, a clinical 102 00:04:01,555 --> 00:04:03,254 AI startup company's perspective, 103 00:04:03,794 --> 00:04:05,955 getting a new tool to be approved by 104 00:04:05,955 --> 00:04:07,574 the FDA is very costly. 105 00:04:08,114 --> 00:04:09,735 And even if you manage 106 00:04:10,034 --> 00:04:12,134 to get your products through the FDA, 107 00:04:12,750 --> 00:04:14,669 you have then make sure that there's a 108 00:04:14,669 --> 00:04:17,149 way to pay for it. And typically in 109 00:04:17,149 --> 00:04:17,810 The US, 110 00:04:18,430 --> 00:04:22,850 that's being done through CPT coding. And oftentimes, 111 00:04:23,229 --> 00:04:25,649 since these AI technologies are new, 112 00:04:26,350 --> 00:04:27,410 there isn't 113 00:04:28,194 --> 00:04:30,055 an applicable CPT code 114 00:04:30,435 --> 00:04:32,055 that could be used for reimbursement. 115 00:04:32,514 --> 00:04:34,355 So now you have to go through the 116 00:04:34,355 --> 00:04:37,735 AMA and apply for a new CPT code. 117 00:04:38,035 --> 00:04:39,394 And let's say you go through the whole 118 00:04:39,394 --> 00:04:39,894 process 119 00:04:40,274 --> 00:04:42,935 and manage to get a new CPT code 120 00:04:43,259 --> 00:04:45,439 to pay for your AI tool, 121 00:04:45,819 --> 00:04:47,979 then you have to go to the insurance 122 00:04:47,979 --> 00:04:50,939 payers and convince them to actually pay for 123 00:04:50,939 --> 00:04:53,339 it. And last but not least, even if 124 00:04:53,339 --> 00:04:55,519 you get some commitment from the payers 125 00:04:55,935 --> 00:04:56,254 to, 126 00:04:56,735 --> 00:04:59,555 pay for your AI tool, there's no guarantee 127 00:04:59,694 --> 00:05:01,475 that it will be reimbursed 128 00:05:01,855 --> 00:05:03,795 at a financially viable level. 129 00:05:04,254 --> 00:05:06,895 However, on the other hand, especially in the 130 00:05:06,895 --> 00:05:09,899 past couple years, I've seen significant traction 131 00:05:10,439 --> 00:05:12,779 in the use of generative AI, 132 00:05:13,160 --> 00:05:13,660 especially 133 00:05:14,040 --> 00:05:16,860 large language models in administrative tests, 134 00:05:17,240 --> 00:05:20,279 such as managing the different steps in revenue 135 00:05:20,279 --> 00:05:21,100 cycle management. 136 00:05:21,964 --> 00:05:24,625 I think this makes sense because these administrative 137 00:05:25,245 --> 00:05:25,745 applications 138 00:05:26,524 --> 00:05:27,665 do not require 139 00:05:28,365 --> 00:05:29,345 FDA approval. 140 00:05:29,884 --> 00:05:31,165 And in many ways, 141 00:05:31,564 --> 00:05:32,384 these administrative 142 00:05:32,925 --> 00:05:33,425 applications, 143 00:05:34,529 --> 00:05:35,349 are easier, 144 00:05:36,050 --> 00:05:37,990 to demonstrate return on investment. 145 00:05:38,769 --> 00:05:40,550 For example, previous studies, 146 00:05:41,169 --> 00:05:42,949 estimated that approximately 147 00:05:43,250 --> 00:05:45,990 30% of health care spending in this country 148 00:05:46,544 --> 00:05:48,084 can be considered waste. 149 00:05:48,865 --> 00:05:52,064 Right now, The US spends about 4,500,000,000,000 150 00:05:52,064 --> 00:05:54,884 per year in health care. So roughly speaking, 151 00:05:55,425 --> 00:05:58,884 the total addressable market is close to $1,000,000,000,000, 152 00:05:59,160 --> 00:06:01,100 meaning there are many opportunities 153 00:06:01,879 --> 00:06:04,220 to do, to use generative AI 154 00:06:04,600 --> 00:06:06,379 to address the inefficiency 155 00:06:06,680 --> 00:06:09,500 and waste in the current health care spending. 156 00:06:10,360 --> 00:06:11,720 That makes a lot of sense, you know. 157 00:06:11,720 --> 00:06:13,420 And I appreciate you going through, 158 00:06:13,879 --> 00:06:15,574 those challenges and headwinds, 159 00:06:16,115 --> 00:06:17,875 because I I think, you know, there's a 160 00:06:17,875 --> 00:06:21,074 lot to consider here, and, certainly, trying to 161 00:06:21,074 --> 00:06:23,955 figure out how, not only to bring the 162 00:06:23,955 --> 00:06:26,275 technology in effectively, but, also, as you mentioned, 163 00:06:26,275 --> 00:06:28,935 the payment methodologies and and the funding available, 164 00:06:29,639 --> 00:06:31,160 for some of these things. And so I 165 00:06:31,160 --> 00:06:33,259 think that's a really, really great watch out. 166 00:06:33,879 --> 00:06:35,800 And along those lines, you know, when you 167 00:06:35,800 --> 00:06:37,980 look at AI and you look at technology 168 00:06:38,199 --> 00:06:40,360 as it grows and evolves, how are you 169 00:06:40,360 --> 00:06:42,920 thinking about growth and continuing to add value, 170 00:06:43,160 --> 00:06:44,540 to the health system overall? 171 00:06:45,295 --> 00:06:48,014 And speaking from the perspective of an academic 172 00:06:48,014 --> 00:06:49,475 medical center, I think 173 00:06:49,855 --> 00:06:53,295 we have to critically evaluate our role in 174 00:06:53,295 --> 00:06:56,895 the rapidly evolving AI landscape and strategically position 175 00:06:56,895 --> 00:06:57,395 ourselves 176 00:06:57,779 --> 00:06:58,439 to maximize 177 00:06:58,740 --> 00:07:00,439 where we can add value. 178 00:07:00,819 --> 00:07:02,740 One thing to keep in mind is the 179 00:07:02,740 --> 00:07:05,060 fact that the majority of care in The 180 00:07:05,060 --> 00:07:06,279 US is delivered 181 00:07:06,660 --> 00:07:08,120 by integrated health systems, 182 00:07:08,580 --> 00:07:10,920 and we will always be the ones performing 183 00:07:11,224 --> 00:07:13,464 the last mile delivery when it comes to 184 00:07:13,464 --> 00:07:14,285 patient care. 185 00:07:14,824 --> 00:07:17,224 It is critical for us to strengthen our 186 00:07:17,224 --> 00:07:17,724 collaborations, 187 00:07:18,584 --> 00:07:21,784 with start ups, industry, and venture capital investment 188 00:07:21,784 --> 00:07:22,284 funds 189 00:07:22,664 --> 00:07:25,305 to ensure that we have a voice in 190 00:07:25,305 --> 00:07:26,604 how these AI 191 00:07:27,009 --> 00:07:29,649 health care products are developed and fine tuned 192 00:07:29,649 --> 00:07:30,709 from day one. 193 00:07:31,250 --> 00:07:32,790 We should also invest 194 00:07:33,170 --> 00:07:35,670 in implementation science and change management 195 00:07:36,209 --> 00:07:39,189 to transform ourselves into world experts 196 00:07:39,564 --> 00:07:42,605 and real world implementation of AI tools in 197 00:07:42,605 --> 00:07:45,805 order to maximize their positive impacts on patients' 198 00:07:45,805 --> 00:07:46,305 care. 199 00:07:47,165 --> 00:07:49,665 I think that's a a really great advice 200 00:07:49,725 --> 00:07:51,324 and certainly makes a lot of sense. 201 00:07:52,089 --> 00:07:54,169 You know? And with AI still being so 202 00:07:54,169 --> 00:07:56,350 new and and changing so quickly, 203 00:07:56,810 --> 00:07:57,389 the capabilities, 204 00:07:57,930 --> 00:07:58,430 rapidly 205 00:07:58,810 --> 00:07:59,310 evolving. 206 00:07:59,769 --> 00:08:02,169 From your perspective, as someone who has spent 207 00:08:02,169 --> 00:08:04,269 a lot of time looking at the technology 208 00:08:04,329 --> 00:08:05,775 side as well as as you mentioned, you 209 00:08:05,775 --> 00:08:08,415 know, the implementation science and change management of 210 00:08:08,415 --> 00:08:10,335 it. What advice do you have for health 211 00:08:10,335 --> 00:08:12,654 systems as they're taking some of these pilot 212 00:08:12,654 --> 00:08:15,295 programs that have worked out well, in in 213 00:08:15,295 --> 00:08:16,175 small groups, 214 00:08:16,654 --> 00:08:18,095 as they're trying to scale it out through 215 00:08:18,095 --> 00:08:18,995 their whole system? 216 00:08:19,970 --> 00:08:21,910 I think this goes back to 217 00:08:22,930 --> 00:08:24,389 a frequently asked question, 218 00:08:25,250 --> 00:08:27,970 when it comes to technology that's not just 219 00:08:27,970 --> 00:08:29,189 confined to AI, 220 00:08:30,050 --> 00:08:32,230 which is to build or buy. 221 00:08:32,725 --> 00:08:35,764 Whenever a new technology comes along, for big 222 00:08:35,764 --> 00:08:38,424 organizations such as an integrated health system, 223 00:08:39,205 --> 00:08:41,764 typically, they have to figure out are they 224 00:08:41,764 --> 00:08:44,965 going to build the new technology themselves or 225 00:08:44,965 --> 00:08:45,465 buy. 226 00:08:46,240 --> 00:08:47,840 Given the pace of, 227 00:08:48,320 --> 00:08:48,820 advancement 228 00:08:49,200 --> 00:08:50,259 when it comes to 229 00:08:50,879 --> 00:08:51,779 AI, especially 230 00:08:52,320 --> 00:08:53,460 generative AI, 231 00:08:54,399 --> 00:08:55,059 I think 232 00:08:55,759 --> 00:08:58,259 my personal opinion is that we should 233 00:08:58,754 --> 00:08:59,894 favor buying. 234 00:09:00,514 --> 00:09:03,394 Just the nature of the technology and the 235 00:09:03,394 --> 00:09:04,934 way that things get built, 236 00:09:05,555 --> 00:09:08,115 I don't think integrated health systems are the 237 00:09:08,115 --> 00:09:08,615 best 238 00:09:09,075 --> 00:09:09,575 actors 239 00:09:10,115 --> 00:09:13,335 in very rapid iteration of these new products. 240 00:09:13,639 --> 00:09:15,980 And if you look at the flourishing marketplace 241 00:09:16,200 --> 00:09:18,220 when it comes to health care AI products, 242 00:09:18,519 --> 00:09:21,320 I really think, you know, either start ups 243 00:09:21,320 --> 00:09:22,059 or industry 244 00:09:22,440 --> 00:09:24,600 will be better suited in building a product 245 00:09:24,600 --> 00:09:25,580 in the first place. 246 00:09:25,960 --> 00:09:26,460 However, 247 00:09:27,794 --> 00:09:30,914 what integrated health systems can do is to 248 00:09:30,914 --> 00:09:34,914 be very actively engaged in in a product 249 00:09:34,914 --> 00:09:37,794 development process in day one. And oftentimes, as 250 00:09:37,794 --> 00:09:38,774 I mentioned earlier, 251 00:09:39,235 --> 00:09:41,014 we are the ones who 252 00:09:41,339 --> 00:09:44,139 are physically interacting with the patients. We are 253 00:09:44,139 --> 00:09:47,179 the ones who really understand best the actual 254 00:09:47,179 --> 00:09:48,879 pain points of health care delivery. 255 00:09:49,500 --> 00:09:52,139 So we are the ultimate customers for these 256 00:09:52,139 --> 00:09:53,600 health care AI tools. 257 00:09:53,934 --> 00:09:55,774 So we should be very involved in the 258 00:09:55,774 --> 00:09:58,675 product development process from day month day one. 259 00:09:59,215 --> 00:10:01,934 So that's my advice number one. And my 260 00:10:01,934 --> 00:10:04,495 advice number two is, in general, if you 261 00:10:04,495 --> 00:10:06,514 look at academic centers or integrated, 262 00:10:06,894 --> 00:10:09,730 health systems, we tend to move a little 263 00:10:09,730 --> 00:10:12,389 bit slower and we tend to be 264 00:10:13,250 --> 00:10:14,549 more on the conservative 265 00:10:14,850 --> 00:10:15,350 side. 266 00:10:15,809 --> 00:10:17,110 I do think that, 267 00:10:17,889 --> 00:10:20,629 we should change our mindset a little bit 268 00:10:20,850 --> 00:10:23,670 and be open to rapid iteration. 269 00:10:24,455 --> 00:10:27,274 And in the sense, you know, we should 270 00:10:27,335 --> 00:10:27,835 aim 271 00:10:28,375 --> 00:10:29,595 to iterate fast, 272 00:10:30,294 --> 00:10:33,995 fail fast if necessary, and pivot even faster. 273 00:10:34,615 --> 00:10:36,554 And the third advice I would give is 274 00:10:37,460 --> 00:10:40,519 every integrated health system should have a robust 275 00:10:40,980 --> 00:10:42,600 AI governance structure. 276 00:10:43,379 --> 00:10:45,700 What I mean is that, I'm sure there's 277 00:10:45,700 --> 00:10:48,279 not a unique problem to Johns Hopkins. 278 00:10:48,820 --> 00:10:49,559 I see 279 00:10:49,914 --> 00:10:52,794 every single integrated health systems being bombarded by 280 00:10:52,794 --> 00:10:55,434 AI request by different vendors. It is a 281 00:10:55,434 --> 00:10:58,875 very chaotic situation right now. So there needs 282 00:10:58,875 --> 00:10:59,774 to be a system 283 00:11:00,235 --> 00:11:01,615 that can really govern 284 00:11:02,049 --> 00:11:05,090 the interaction between AI vendors and integrated health 285 00:11:05,090 --> 00:11:07,269 system. And the ultimate goal is really, 286 00:11:07,970 --> 00:11:11,029 so that we can pick AI tools that 287 00:11:11,090 --> 00:11:13,330 are feasible, that make sense in the real 288 00:11:13,330 --> 00:11:14,470 world, and ultimately, 289 00:11:14,929 --> 00:11:17,190 are safe and efficacious for patients. 290 00:11:17,644 --> 00:11:20,524 So, this is something that Johns Hopkins Medicine 291 00:11:20,524 --> 00:11:21,665 started doing recently, 292 00:11:22,365 --> 00:11:25,165 essentially starting about a year ago. The health 293 00:11:25,165 --> 00:11:26,065 system leadership 294 00:11:26,605 --> 00:11:27,904 recognizes the need 295 00:11:28,445 --> 00:11:29,264 to really, 296 00:11:29,929 --> 00:11:33,610 evaluate and implement AI technology at scale in 297 00:11:33,610 --> 00:11:35,470 a responsible and safe manner. 298 00:11:36,089 --> 00:11:37,950 And in order to achieve that goal, 299 00:11:38,409 --> 00:11:41,690 a decision was made to establish an AI 300 00:11:41,690 --> 00:11:42,750 governance structure 301 00:11:43,274 --> 00:11:45,595 to really govern and figure out what is 302 00:11:45,595 --> 00:11:47,454 the optimal best practices 303 00:11:48,074 --> 00:11:50,735 when it comes to interacting with AI vendors. 304 00:11:50,875 --> 00:11:53,514 So for example, now at Johns Hopkins, if 305 00:11:53,514 --> 00:11:56,720 you were an AI vendor that is interested 306 00:11:57,179 --> 00:12:00,000 in selling to or working with Johns Hopkins, 307 00:12:00,459 --> 00:12:03,659 there is a standardized intake process that you 308 00:12:03,659 --> 00:12:04,799 you have to go through. 309 00:12:05,419 --> 00:12:07,759 There will be a set of standardized questions 310 00:12:08,139 --> 00:12:10,720 that cover things like the model card, 311 00:12:11,115 --> 00:12:12,575 the, data provenance, 312 00:12:13,355 --> 00:12:16,414 the validation process. And after the initial intake, 313 00:12:16,955 --> 00:12:18,975 depending on the nature of the application, 314 00:12:19,434 --> 00:12:22,095 the application will then be sent to specific, 315 00:12:23,115 --> 00:12:23,615 subcommittees 316 00:12:24,309 --> 00:12:27,129 that are composed of a variety of stakeholders, 317 00:12:27,430 --> 00:12:28,970 including physician leaders 318 00:12:29,430 --> 00:12:32,149 who then do a deeper dive into the 319 00:12:32,149 --> 00:12:34,090 merit and safety of these applications. 320 00:12:34,710 --> 00:12:37,430 And so every vendor now is expected to 321 00:12:37,430 --> 00:12:38,649 go through the same process 322 00:12:39,024 --> 00:12:41,904 being evaluated the same way, and we will 323 00:12:41,904 --> 00:12:45,105 also provide adequate feedback to the vendors, at 324 00:12:45,105 --> 00:12:47,745 the same time. It's not meant to be 325 00:12:47,745 --> 00:12:48,485 a burdensome, 326 00:12:49,024 --> 00:12:49,524 process, 327 00:12:49,904 --> 00:12:52,519 but, really, the ultimate goal is to establish 328 00:12:52,579 --> 00:12:55,720 something that's standardized that can maximize the safety 329 00:12:56,100 --> 00:12:58,579 and efficacy of these AI tools for patient 330 00:12:58,579 --> 00:12:59,079 care. 331 00:12:59,860 --> 00:13:01,699 I love that. I think that's such a 332 00:13:01,699 --> 00:13:03,379 a smart way of going about it to 333 00:13:03,379 --> 00:13:04,199 really organize, 334 00:13:05,174 --> 00:13:06,934 all the noise. As you mentioned, so many 335 00:13:06,934 --> 00:13:09,735 different pitches coming through and vendors, you know, 336 00:13:09,735 --> 00:13:12,154 wanting to connect with members of the organization, 337 00:13:12,455 --> 00:13:13,595 clinical, administrative, 338 00:13:14,615 --> 00:13:16,615 across the board. And and so I I 339 00:13:16,615 --> 00:13:18,855 think that is truly helpful to understand how 340 00:13:18,855 --> 00:13:20,597 you're dealing with that at Johns Hopkins, and, 341 00:13:21,299 --> 00:13:23,139 I I can imagine it will continue to 342 00:13:23,139 --> 00:13:23,960 evolve over time. 343 00:13:25,620 --> 00:13:27,299 And and speaking of that, where do you 344 00:13:27,299 --> 00:13:29,539 see some of the best opportunities for growth 345 00:13:29,539 --> 00:13:30,360 in the future? 346 00:13:31,299 --> 00:13:34,340 One area that I'm particularly excited about is 347 00:13:34,340 --> 00:13:35,080 the intersection 348 00:13:35,460 --> 00:13:37,274 of AI and oculomix. 349 00:13:38,054 --> 00:13:40,075 So as a some background information, 350 00:13:41,095 --> 00:13:41,595 oculomix 351 00:13:42,054 --> 00:13:44,875 is a field of study that connects findings 352 00:13:44,934 --> 00:13:45,675 or biomarkers 353 00:13:46,455 --> 00:13:48,774 from the retina, which is the back of 354 00:13:48,774 --> 00:13:51,434 the eye, with systemic health states. 355 00:13:51,889 --> 00:13:54,850 For example, if we use deep learning, which 356 00:13:54,850 --> 00:13:57,889 is the current cutting edge AI technique for 357 00:13:57,889 --> 00:13:58,710 image analysis 358 00:13:59,410 --> 00:14:01,590 to analyze a retina photograph, 359 00:14:02,289 --> 00:14:04,629 we can now predict someone's age, 360 00:14:05,024 --> 00:14:07,924 biological sex, smoking status, blood pressure, 361 00:14:08,464 --> 00:14:12,404 cardiovascular risks, presence of dementia, and kidney status. 362 00:14:13,184 --> 00:14:15,904 This is possible because the retina is the 363 00:14:15,904 --> 00:14:16,804 only place 364 00:14:17,184 --> 00:14:18,565 in the entire body 365 00:14:19,049 --> 00:14:21,690 where we can image both blood vessels and 366 00:14:21,690 --> 00:14:23,549 brain tissue at the same time 367 00:14:23,929 --> 00:14:24,429 noninvasively. 368 00:14:25,610 --> 00:14:27,850 This is no longer science fiction, although it 369 00:14:27,850 --> 00:14:29,230 sounds like science fiction. 370 00:14:29,850 --> 00:14:33,070 In fact, there are now multiple startup companies 371 00:14:33,715 --> 00:14:35,575 that are actively involved in oculomix. 372 00:14:36,355 --> 00:14:38,595 This approach is very exciting to me because 373 00:14:38,595 --> 00:14:39,894 it has the potential 374 00:14:40,434 --> 00:14:41,335 to upend 375 00:14:41,794 --> 00:14:44,595 medicine as we know it. So imagine in 376 00:14:44,595 --> 00:14:46,294 the not so distant future, 377 00:14:46,759 --> 00:14:49,100 Laura, you can get your retina 378 00:14:49,480 --> 00:14:52,139 image at your local pharmacy or supermarket. 379 00:14:52,840 --> 00:14:54,940 Within minutes, your retina photographs 380 00:14:55,559 --> 00:14:58,840 will be analyzed by several different AI models, 381 00:14:59,080 --> 00:15:00,620 hosted in in the cloud 382 00:15:00,934 --> 00:15:03,414 that would then give you a readout for 383 00:15:03,414 --> 00:15:06,394 your systemic health states for multiple organs. 384 00:15:07,254 --> 00:15:10,074 As you can imagine, this kind of community 385 00:15:10,134 --> 00:15:12,875 screening is obviously much more convenient 386 00:15:13,495 --> 00:15:14,714 and cost effective 387 00:15:15,230 --> 00:15:17,730 than getting a brain MRI scan 388 00:15:18,110 --> 00:15:20,350 or a CT scan of your heart and 389 00:15:20,350 --> 00:15:23,009 has the potential to reach millions of patients 390 00:15:23,470 --> 00:15:25,649 who are not even aware of the systemic 391 00:15:25,709 --> 00:15:27,254 health problems to begin with. 392 00:15:27,735 --> 00:15:28,634 And, historically, 393 00:15:29,174 --> 00:15:30,875 retina specialists like myself 394 00:15:31,495 --> 00:15:34,695 are considered super specialists, so you only get 395 00:15:34,695 --> 00:15:38,075 to see a retina specialist after multiple referrals. 396 00:15:38,535 --> 00:15:41,174 But now we are really flipping this health 397 00:15:41,174 --> 00:15:43,840 care journey on its head and are using 398 00:15:43,980 --> 00:15:46,540 the retina literally as a window to your 399 00:15:46,540 --> 00:15:49,600 body as a form of preprimary care. 400 00:15:49,980 --> 00:15:50,960 This potentially, 401 00:15:51,980 --> 00:15:55,440 paradigm shifting delivery model could have huge implications 402 00:15:55,580 --> 00:15:57,120 for the entire health care industry, 403 00:15:57,835 --> 00:15:59,934 especially for integrated health systems. 404 00:16:00,955 --> 00:16:02,955 Wow. That's amazing to hear and, you know, 405 00:16:02,955 --> 00:16:05,274 really, truly an exciting field of study, as 406 00:16:05,274 --> 00:16:05,835 you mentioned. 407 00:16:06,315 --> 00:16:08,715 You know, having that opportunity not only to 408 00:16:08,715 --> 00:16:10,335 see on the clinical side 409 00:16:11,389 --> 00:16:14,029 a very, very clear, return on investment for 410 00:16:14,029 --> 00:16:16,910 patients who who, you know, want to learn 411 00:16:16,910 --> 00:16:18,509 more about what's going on with their bodies, 412 00:16:18,509 --> 00:16:19,809 but also just the opportunity 413 00:16:20,190 --> 00:16:23,384 to democratize the that knowledge and information and 414 00:16:23,705 --> 00:16:24,684 ability to, 415 00:16:25,225 --> 00:16:28,125 have access to these types of screenings. So 416 00:16:28,345 --> 00:16:30,184 I I think that's really, really cool. And, 417 00:16:30,345 --> 00:16:32,345 again, as you mentioned, just another example of 418 00:16:32,345 --> 00:16:34,745 how the technology can make a big difference, 419 00:16:34,985 --> 00:16:36,745 within the health care space and and the 420 00:16:36,745 --> 00:16:38,125 health of communities overall. 421 00:16:39,230 --> 00:16:40,750 Yeah. I definitely agree. 422 00:16:41,710 --> 00:16:44,350 When I started, you know, getting involved in 423 00:16:44,350 --> 00:16:46,129 the field of AI research 424 00:16:46,750 --> 00:16:49,389 about eight years ago, you know, things were 425 00:16:49,389 --> 00:16:52,384 already advancing really fast. And at that 426 00:16:52,764 --> 00:16:54,545 time, retina and radiology, 427 00:16:55,644 --> 00:16:58,225 were at the forefront of this revolution because 428 00:16:58,524 --> 00:17:01,725 the current iteration of, AI, as I mentioned 429 00:17:01,725 --> 00:17:05,025 earlier, the cutting edge, technique is deep learning. 430 00:17:05,400 --> 00:17:07,320 It is very good at two things. One 431 00:17:07,320 --> 00:17:09,980 is image analysis, and the second one 432 00:17:10,279 --> 00:17:13,500 is in understanding human language. But the ability 433 00:17:13,720 --> 00:17:15,740 to really analyze images 434 00:17:16,680 --> 00:17:17,420 at a level 435 00:17:17,720 --> 00:17:19,019 not possible before 436 00:17:19,515 --> 00:17:21,674 first came to our attention about eight years 437 00:17:21,674 --> 00:17:23,914 ago. So if you look at the trajectory 438 00:17:23,914 --> 00:17:26,974 of medical AI, the medical fields 439 00:17:27,434 --> 00:17:28,734 that were most, 440 00:17:29,434 --> 00:17:30,734 reliant on images, 441 00:17:31,595 --> 00:17:34,410 were the ones who really got into, this 442 00:17:34,410 --> 00:17:37,769 AI revolution early on, and these fields are 443 00:17:37,769 --> 00:17:39,230 radiology and ophthalmology. 444 00:17:39,609 --> 00:17:42,089 So we started doing all this research even 445 00:17:42,089 --> 00:17:42,589 before 446 00:17:43,210 --> 00:17:46,250 the relatively recent boom in large language models, 447 00:17:46,250 --> 00:17:48,029 which started about two years ago. 448 00:17:49,005 --> 00:17:50,924 It's amazing to hear. And it's just really 449 00:17:50,924 --> 00:17:53,404 cool to know and see that history, and 450 00:17:53,404 --> 00:17:55,725 I can imagine a continued opportunity for the 451 00:17:55,725 --> 00:17:58,045 evolution of the field in really, really meaningful 452 00:17:58,045 --> 00:17:59,884 ways. Doctor Liu, thank you so much for 453 00:17:59,884 --> 00:18:01,650 joining us on the podcast today. This has 454 00:18:01,650 --> 00:18:04,210 been so inspiring and a fun conversation for 455 00:18:04,210 --> 00:18:06,529 me, and I I look forward to connecting 456 00:18:06,529 --> 00:18:09,170 with you again soon and, just continuing on, 457 00:18:09,410 --> 00:18:11,809 the journey of the progress along along these 458 00:18:11,809 --> 00:18:13,330 lines. So thank you so much for your 459 00:18:13,330 --> 00:18:15,432 time today. Same here, Laura. It's been a 460 00:18:15,432 --> 00:18:17,752 pleasure. I look forward to chatting with you 461 00:18:17,752 --> 00:18:18,492 next time.