Machine Learning for MDs Weekly Digest
What’s New in ML for MDs
Welcome to the ML for MDs Newsletter. The mission of ML for MDs is to connect physicians interested in machine learning. This newsletter provides the most relevant news, journal articles, and jobs at the intersection of medicine and machine learning.
Weekly watching
This week’s video is a 4 minute video on tradeoffs in the design of machine learning models.
Weekly summary
Here’s a table of some short courses (days to a few months) on Healthcare AI. These are not endorsements; they are courses that come up frequently in searches for healthcare AI courses.
There seem to be two major types of courses:
- Courses aimed at programmers to give them some background in healthcare in a project-based way. These also tend to be in the hundreds of dollars.
- Courses aimed at physicians to give them a broad overview of AI. These tend to be in the thousands of dollars for less course time (shocking, I know!)
There are several free courses online listed on MLforMDs.com, or in the Slack #resources channel.
There’s also a Deep Learning Specialization on Coursera with Andrew Ng; I’ve taken some of his courses (but not all!) and enjoyed them. However, it’s not healthcare-specific.
Please let me know if you’ve taken other courses you’d recommend (or would recommend avoiding).
| Program | Institution | Cost | Target audience | Pre-reqs | Level (intro/intermed/adv) | Math | Programming | Length | Online/in person | Degree/certificate |
| AI for Healthcare: Concepts and Applications | Harvard | $2,600 | This online program is designed for senior managers and executives who are responsible for developing and implementing AI strategy in their organizations and are looking to understand AI, its current state of the art, and future. | None | Intro | None | None | 4 days | Online, synchronous | Certificate of Participation and Continuing Education Units |
| Designing and Implementing AI Solutions for Healthcare | Harvard | $2,150 | “Changemakers” at healthcare companies | None | Intro | None | None | 3 weeks, 10-12h/week | Online, synchronous | Unclear |
| Artificial Intelligence in Healthcare | Stanford | $79/month via Coursera | This specialization is designed for both healthcare providers and computer science professionals, offering insights to facilitate collaboration between the disciplines. | None | Intro | None | None | About 55h plus a capstone project | Online, asynchronous | Stanford Certificate of Achievement in Artificial Intelligence in Healthcare |
| Artificial Intelligence in Healthcare Short Course | MIT | $2,800 | Run by the Executive Education program at Sloan Business School – focused on healthcare leaders | None | Intro | None | None | 6 weeks, 6-8h/week | Online, asynchronous | Digital MIT Sloan certificate |
| AI for Healthcare | Udacity | $399/month for about 4 months | Data scientists/programmers interested in learning about project-based healthcare applications | Intermediate Python, Intermediate Statistics | Intermediate | None | Yes | 4 months, 15h/week | online, asynchronous | Udacity Nanodegree |
| AI for Medicine Specialization | Deeplearning.ai via Coursera | $49/month for about 3 months | Data scientists/programmers interested in learning about project-based healthcare applications | “You can program in Python and are comfortable with statistics and probability.” | Intermediate | None | Yes | 3 months at 7 hours/week | online, asynchronous | Coursera Specialization |
Fun Facts
- The CDC’s supercomputer 7600, the fastest supercomputer in 1975, cost $5 million at the time (about $32 million in today’s money).
- The iPhone 4 had the same performance characteristics but cost only $400
- Machine learning solutions deployed by one police department reduced the murder rate by 35% and robberies by 20% with a forecasting mechanism that optimized the deployment of police units. But these have increasingly been criticized for racial and socioeconomic bias.
This Week’s Top Stories
- Cortical labs, which is trying to grow brain cells with computer chips, raised $10million from Hong Kong’s richest person.
- As AI advances, privacy issues will increase. In the non-medical world, this is called PII (Personally Identifiable Information) rather than PHI. Laws about PII vary by country and even by state. This article gives a nice summary of the history of PII and its many forms.
- Check out this look at health AI from an overall/clinical/business standpoint. I was surprised to see that the amount of time in the EHR is pretty similar across specialties, though the number of encounters for that time varied wildly. To me that signifies that there is a painful threshold around 20 hours a week at which point physicians start hiring extenders.
Community News
- If you haven’t introduced yourself, please do so under the #intros channel.
Thanks for being a part of this community! As always, please let me know if you have questions/ideas/feedback.
Sarah
Sarah Gebauer, MD