Medical AI can detect the racial identity of patients from x-rays. This is extremely concerning, and raises urgent questions about how we test medical AI systems.
Docs are ROCs: a simple fix for a “methodologically indefensible” practice in medical AI studies.
The way we currently report human performance systematically underestimates it, making AI look better than it is.
CT scanning is just awful for diagnosing Covid-19
Reports that CT scanning may be better than PCR testing for covid-19 are flawed and almost certainly wrong.
Improving Medical AI Safety by Addressing Hidden Stratification
Medical AI testing is unsafe, but addressing hidden stratification may be a way to prevent harm, without upending the current regulatory environment.
AI competitions don’t produce useful models
Ai competitions are fun, community building, talent scouting, brand promoting, and attention grabbing. But competitions are not intended to develop useful models.
The best medical AI research (that you probably haven’t heard of)
I discuss a piece of medical AI research that has not received much attention, but actually did a proper clinical trial!
Half a million x-rays! First impressions of the Stanford and MIT chest x-ray datasets
My first impressions of these datasets. How do they measure up, and how useful might they be?
Explain yourself, machine. Producing simple text descriptions for AI interpretability.
Humans explain their decisions with words. In our latest work, we suggest AI systems should do the same.
The unreasonable usefulness of deep learning in medical image datasets
Medical data is horrible to work with, but deep learning can quickly and efficiently solve many of these problems.
Post-doc and PhD positions for deep and reinforcement learning in medical images
Our team has post-doc and PhD positions available, so come to unexpectedly great Adelaide!