Dr. Chris Lovejoy is a medical doctor who has spent the last several years exploring data science. He has a master’s degree in machine learning at UCL, and he currently works as a data scientist. He’s been involved with several AI for healthcare projects, including a model to predict health deterioration, and a BMJ systematic review looking at the quality of AI studies.
On his blog and YouTube channel, he shares his thoughts and experiences of applying machine learning to healthcare. Most recently, one of his fun coding projects where he created his own alternative to the YouTube algorithm for finding interesting videos went viral.
Summary
In this episode, Dr. Lovejoy starts by delving into why he decided to code up his own YouTube videos recommendation algorithm. We then discuss the state of AI in medical research and explore what could be the gold standard of reporting results on the use of AI in medical studies. Chris explains how explainthispaper.com started and how is it going and shares his experiences in building a engaged community interested in Healthcare + AI.
Key Questions
9:36.67: Tell us a little more about how you as a medical doctor got interested in Data Science and Machine Learning, AI and what made you sort of pursue this path?
12:22.68: …are you now more of a data scientist than a medical doctor…?
20.58.74: …is there a gold standard of how these studies (AI in medicine) must be reported?
38:28.73: Someone who’s listening to this conversation, if they’re interested to contribute to writing a summary (of a paper @ explainthispaper.com) what should they do?
39:51.25: So, what have been your learnings from building and engaging with this community of people who are sort of interested in this intersection of healthcare and AI?
41:27.61: Doctors who can understand technology and work with it or a technologist who can understand medicine. Which of those flavors would work better for problems in the healthcare space?
Quotable Quotes
3:44.20: I just had the idea that maybe I could make an alternative YouTube algorithm. It would recommend me videos that are a bit more tailored to what I’m interested in and essentially just kind of lead to a higher quality of videos on YouTube.
14:05.56: When you’re building out your data science skills, you want to be maintaining your clinical skills and when you’re building clinical skills, you want to be maintaining your data science skills, but it’s very hard to build both at the same time.
19.52.12: We’re leaning towards trying to prove that AI is better than humans and trying to kind of get a headline that this new algorithm is beating doctors or whatever.
23.00.81: …important considerations are, how are we going to implement the algorithms in the clinical workflows? How are we going to monitor that performance over time? How are we going to make sure that we are becoming aware if performance changes and can kind of intervene in that?
24:51.79: A good clinical performance metric does not lead directly to a good clinical outcome. So sometimes our model will perform a task well, but actually when you implement it in a clinical setting, then it might not have a positive impact or may even have a negative impact.
36:09.09: We want to focus on the papers that either present something new, that’s not really maybe covered in other summaries or it’s something new in the field.
42:27.22: As a technologist I think there’s so many things about clinical medicine that are different to other industries and that are quite unique to medicine which makes it difficult to understand if you haven’t worked as a doctor.
Notable Mentions
11:16.55: Python for Data Science course
11:18.74: Andrew Ng’s Machine Learning course
32:58.98: explainthispaper.com
Connect
Website: https://chrislovejoy.me/
Twitter: @ChrisLovejoy_
LinkedIn: Dr Chris Lovejoy
The You+AI Vodcast
A companion video segment full of fun and candid moments.
Check it out here!