Motivation for the Program
There is an exploding interest in the use of AI for medical application and more generally in healthcare. However, there are (at least) two common misconceptions that we need to address. The first is that AI is simply magic and that its use will end all the issues we face. The second is that all we really need is a better set of algorithms. The pictures shown on this page hopefully help frame the current reality. (Both are AI-generated using Microsoft Copilot.) The first shows a car with its hood open and the engine visible. This can be thought of a good illustration of the role of AI tools in modern medical software. The AI is a component of the car, perhaps an important component such as the engine. However, our users drive a car not an engine, and as such, and ultimately the solution to the needs of our user is going to be a complete software tool as opposed to just a better neural network. This is the motivation behind structuring this program as a combined medical software/medical AI course. We need to understand about AI (engine), and how to design and optimize it for the task at hand. But we must remember that ultimately our users use systems, which consist of multiple components, and, importantly, that the FDA and other regulatory agencies, review systems not algorithms. Furthermore, issues such as bias, security, usability, are often better addressed as the system (car) level and not necessarily at the AI (engine) level.
The second picture is that of a radiologist looking at a workstation showing a chest x-ray. The question here is what is the system? Is the system the software (plus the computer), or does the system also include the radiologist? From the perspective of the patient whose chest x-ray is being viewed, the system in fact includes the radiologist, as the diagnosis will be the output of the combined effort of the human plus whatever “AI” is running inside our software. Hence our task does not end with the design of our tool, but we must consider how this integrates with the human component of medicine. As in the previous example, issues with our software (e.g., safety risk) may also be better addressed at the overall system (human + software) level as opposed to the software or the AI level.
Program Overview
The certificate program consists of four modules. In the first we present an introduction to medical software. This includes the regulatory aspects, foundational management tools such as quality systems and risk management, and a description of the actual lifecycle process of designing, implementing, and testing the tools. At this stage, we focus on traditional software without any AI components.
The second module presents a rapid overview of artificial intelligence or machine learning (in our use the two terms are basically synonymous) and relate this to classical probability theory which lies at the heart of these algorithms. We discuss both classical machine learning methods, neural networks, and present an introduction to the reason for all the current excitement in the field, “generative AI” techniques.
The third module of the program aims to integrate the first two and to think about how AI changes the software process, both from a regulatory and an engineering aspect. In addition, the use of substantial amounts of data to create the algorithms means that we need to think both about data organization and data privacy. Finally, the excitement around AI means that we also have new legislation/regulations that, while non-medical in origin (e.g. the EU AI Act), affect the creation of medical software tools.
The concluding fourth module focuses on applications. First, we discuss imaging, which represents close to 75% of all FDA-cleared AI tools on the market. Next, we discuss the issues involved in the creation of modern clinical decision support tools. After this, we take a trip to Africa to discuss the creation and use of such tools in a global health/low resource setting. The last week concludes the program with vignettes ranging from the use of AI to improve healthcare delivery, to multimodal AI, to emerging applications such as pharmaco-genomics.
Program Structure
The program is structured in a 4 week on/1 week off format. Each week, the students will be given 60-75 minutes of video material to review and then be asked to complete a short 12-15 question multiple choice quiz. We will then discuss the material in weekly live zoom sessions scheduled appropriately (i.e., there will be more than one session) to allow students from multiple timezones to participate. (In 2024, we had two sessions, one at 12 pm EST, and one at 8 pm EST.) Students will be to select whichever session was most convenient on any given week. The course director, the instructors for the week, and guest experts from industry will join the zoom sessions to answer questions and to offer additional thoughts.
Weekly Schedule:
- Videos released on Wednesdays. (e.g., 1/15/2025)
- Quizzes due on Mondays. (e.g., 1/20/2025)
- Zoom review sessions on Tuesdays. (e.g., 1/21/2025)
Dates
Each module runs for four weeks from a Wednesday to a Tuesday.
Module 1: 1/15/2025 – 2/21/2025
Module 2: 1/19/2025 – 3/18/2025
Module 3: 3/26/2025 – 4/22/2025
Module 4: 4/30/2025 – 5/27/2025