Have you ever had an idea that you just had to make real? No matter what it took… no matter what obstacles were in your way… no matter how many times people told you no… you just couldn’t stop until it existed? Well, this is one of those stories. It begins with an idea in 1988 and leads to the first-ever autonomous AI to be approved by the FDA for diagnosis without physician input.
Dr. Michael Abramoff, MD, Ph.D. is the Founder and Executive Chairman of Digital Diagnostics, the autonomous AI diagnostics company which was the first in any field of medicine to get FDA authorization for an autonomous AI. Dr. Abramoff is a neuroscientist, a practicing physician, and holds a Ph.D. in Artificial Intelligence and Machine Learning. In 1988, Michael was working on artificial intelligence during his residency and began to think a computer could diagnose diabetic retinopathy. Given the technology available at the time, this idea may have been a bit of a stretch. Still, Michael set out to prove it could be done.
Joining him in this interview is Seth Rainford, the President and COO at Digital Diagnostics. Seth focuses on expanding market opportunities and driving operational excellence within the company. He brings more than a decade of executive experience to Digital Diagnostics including the successful management of large-scale P&L’s, strong organic & inorganic business development expertise, as well as complex multi-site operations leadership within the healthcare industry.
In this episode, we talk with Dr. Abramov and Seth about the 30-year journey that led to the founding of Digital Diagnostics, and the first-ever FDA-approved Autonomous AI in healthcare. Plus, we explore the challenges they continue to work through as they commercialize their product to support organizations looking to win in value-based care!
01:30 Introduction to Dr. Abramoff and Seth Rainford and how the first-ever autonomous AI solution became FDA-approved for diagnosis without physician input
03:30 The scalability of Artificial Intelligence in healthcare and the recent failure of IBM Watson Health
06:00 “We are at an inflection point with AI…specifically with Autonomous AI.”
06:30 The parallel paths between AI and the discovery of DNA and its eventual use in the courtroom.
07:45 Why should we limit diagnosis to human cognition when autonomous AI has been proven to be safe and effective?
08:45 An overview of the history of AI, from advancements in neuroscience and sensory processing, ML, artificial neural networks, to autonomous AI in healthcare.
10:45 Where IBM Watson failed – it started with “glamour AI” (i.e. winning at Jeopardy) instead of trying to solve problems in healthcare
12:00 Most of what we hear about in healthcare is assistive AI — not autonomous AI.
13:20 There is no need for human oversight in autonomous AI for making FDA-approved diagnoses in healthcare.
15:15 Referencing a recent NEJM Catalyst Op-Ed that criticizes autonomous AI in healthcare
16:30 Lessons learned from the challenges of assistive AI and how the develop of a completely autonomous AI solution started with FDA approval
18:30 “In considering the best ways to improve population health outcomes, we must include autonomous AI.”
19:00 Humans are not necessarily better than AI when it comes to diagnosis of diabetic retinopathy
19:20 Referencing NEJM study using assistive AI diagnosis of breast cancer and how radiologist involvement with AI didn’t improve outcomes
23:00 Recent CMMI focus to advance health equity in value-based care
23:45 “Diabetic Retinopathy is the main cause of blindness and this is disproportionately impacting minorities and rural populations due to lack of access to care.”
25:00 The importance of the diabetic eye exam and how to make testing more accessible through autonomous AI
26:10 Democratizing access to diabetic retinopathy exams in underserved areas through scalable platforms and innovative partnerships
29:30 Should Autonomous AI systems replace traditional testing approaches as a standard of care?
30:30 Diagnosability that is comparable between all population segments
31:30 The difference between “screening” and “early detection”
32:00 Using telemedicine to overread diagnostic findings in autonomous AI will cause unnecessary delays and avoid real-time referral and immediate treatment
34:00 Testing autonomous AI at the point-of-care in the primary care setting demonstrates effectiveness and scalability potential
35:00 “Our AI platform is eminently scalable. Getting a point-of-care diagnostic in real-time with a low-skilled operator democratizes access at one-third of the cost.”
36:00 Unlocking value for all stakeholders – payers, patients, and providers
39:00 How does early diagnosis and detection of diabetic retinopathy supports risk bearing entities with risk adjustment coding and closing care gaps?
42:40 The simplified CPT coding for early diagnosis and detection of diabetic retinopathy with Autonomous AI
45:00 How autonomous AI improve physician productivity and effectiveness, while also reducing burnout and suffering in the physician workforce
47:30 The concern of algorithmic bias in AI and how that may contribute to health inequities and disparities in care
49:30 How clinical trials and AI training design mitigate risk of implicit biases within AI algorithms
50:40 Referencing “Coded Bias” (show on Netflix that investigates the bias in algorithms using facial recognition technology)
53:30 How does black-box AI versus biomarker-based AI compare in bias mitigation and patient safety?
55:30 The future applications of autonomous AI in value-based care (e.g. emerging use cases and the potential intersection of precision health and genomics)