Own the Unknown? with Matthew Lungren

Welcome to the first January issue of Further's Own the Unknown? LinkedIn newsletter, which means it is time to introduce you to a new thought leader. Twice monthly, we'll share some of the knowledge we've gained from following, reading, and interviewing some of the most insightful and influential thought leaders on LinkedIn.

This month we’ll be discussing the thought leadership of Matthew Lungren MD MPH , Chief Scientific Officer for 微软 Health & Life Sciences... although that is just one of several roles.?

Keith McCormick will be interviewing Matthew tomorrow, January 8th, at 1 pm ET. You will be able to find us directly in Further’s LinkedIn feed. We recommend following Further now, and then going directly to the feed for the event. The recording will immediately get saved to the feed, but we encourage you to join us live so that you can ask questions during the event.

Matthew's scientific work has led to more than 150 publications, including work on multi-modal data fusion models for healthcare applications, new computer vision and natural language processing approaches for healthcare specific domains, opportunistic screening with machine learning for public health applications, open medical data as public good, prospective clinical trials for clinical AI translation, and application of generative AI in healthcare. He has served as advisor for early stage startups and large fortune-500 companies on healthcare AI technology development and go-to-market strategy. Dr. Lungren's work has been featured in national news outlets such as NPR, Vice News, and Scientific American, and he regularly speaks at national and international scientific meetings on the topic of AI in healthcare.

As a physician and clinical machine learning researcher, he maintains a part-time clinical practice at UCSF while also continuing his research and teaching roles as adjunct professor at 美国斯坦福大学 . Matthew is a prolific contributor to content online. The YouTube channel for the Stanford Center for Artificial Intelligence in Medicine and Imaging (AIMI) is an outstanding resource, including a recent keynote presentation by Matthew. Noteworthy mention from the keynote was his strong recommendation of reading The AI Revolution in Medicine.

Central to our discussion will be his excellent one-hour LinkedIn Learning course, An Introduction to How Generative AI Will Transform Healthcare. His Coursera course is also very highly regarded. This interview is an option for a candid discussion that is more biographical in nature.

To orient ourselves to this extensive body of work, we’ll focus on some of the most important use cases for AI in healthcare. We’ll discuss just three, but this is such a rapidly growing field that it represents the merest sampling of the exciting applications of Medical AI.

Clinical Trials

Matthew’s combined roles at Microsoft, as an academic, as a researcher, and as a practicing physician give him a front row seat to some of the most promising areas. Some remarkable facts from his course:

Nearly half of the 500,000 clinical trials for new therapies fail to reach enrollment targets? And at the same time, only 3% of US patients eligible for trials successfully find them. This inefficiency leads to billions of dollars in waste and prevents access to potentially new cures for patients. So what's behind this? Well, it turns out that matching patients to clinical trials is incredibly difficult. There could be dozens of trials per condition, and each trial has unique, complicated eligibility criteria, and it's all written up in a very complicated medical jargon. And at the same time, clinical trial researchers struggle to find participants often because it requires a deep analysis of candidates medical data just to determine if they qualify.?

Generative AI promises to help with this matching process. In the course, Matthew describes how Microsoft and Providence Health worked on this problem. Once you are alerted to the possibilities, you start to see them everywhere like this TEDx talk about using AI to identify sites for a phase three vaccine trial.??

Medical Transcription

Seeking to use technology to address medical transcription is not new. In the more biographical interview, mentioned earlier, he describes using NLP and Support Vector Machines in the years prior to the rise of Deep Learning. We’ll be sure to ask about this because our own team at Further includes NLP experts, and we often draw upon that expertise to supplement our use of Gen AI.

Matthew describes the challenge in his LinkedIn Learning course:

For decades, the healthcare system has grappled with the challenge of paperwork. Today on average, healthcare providers spend about 15 1/2 hours per week on administrative tasks like documenting electronic health records. In fact, the American Medical Association recently reported that 2/3 of physicians attribute burnout to administrative tasks, and as a result, half are considering leaving medicine altogether. So clearly there is a crisis and it's past time that we find approaches that can help.

Part of his role at Microsoft is to work with a company called Nuance on technologies to address this need. Getting an update on the state of the art will be part of our discussion.

Medical Imaging

As a radiologist, imaging has always been an important part of Matthew’s work. A talk of Matthew’s that is now a few years old gives a good overview of this world. A powerful use case that emerges when one researches this is using AI instead of medical contrast dye. For many, computer vision would seem to place the doctor and AI in competition, with AI attempting to identify abnormalities directly. The contrast dye use case, instead, prepares the image for the radiologist. This Oxford University news article describes the problem of traditional contrast agents.

Imagine you are a medical doctor, faced with a patient with suspected heart disease for symptoms such as chest pain, tightness, or shortness of breath. One way to find out what is happening, and help guide patient prognosis, is to do a cardiovascular MRI scan to look into any heart muscle abnormalities. The scan involves injecting a ‘contrast agent’ (a dye that will improve image contrast and show up scars on images) into a vein in the patient. Contrast-enhanced MRI has been the clinical standard to provide clear scar images, but it’s painful, and makes already expensive MRI scans even more so.

Perhaps surprisingly, this work also began before the famous 2012 victory of Deep Learning in the ImageNet competition that prompted deep neural networks to enter the mainstream. Since Matthew’s work in this area predates that time period, we’ll be sure to ask him how these use cases have evolved from before 2012, to the period after, and now since ChatGPT 4.0.

Looking Ahead

As we look ahead to February, we’re excited to continue our conversations with AI ethicist Olivia Gambelin .

Also, multiple members of the Further team will be leading masterclasses at TDWI Las Vegas, where they will be joined by our recent guest, Donald Farmer . You can read more about the conversation he and Keith McCormick had back in November here. You can find a description of the masterclass Hands-On Introduction to Customizing Large Language Models (LLMs) by Further's Kristy Hollingshead and Nicolas Decavel-Bueff here. Keith will be leading AI?and Deep Learning: Techniques and Use Cases for Profitable Applications as well as the Data Science Bootcamp series.