Curt Langlotz Profile picture
Director of @StanfordAIMI center, radiologist @RSNA board member & informatics geek, @StanfordBMI alum, hacker at heart, author of https://t.co/baCeP17pU5
Sep 4, 2021 4 tweets 6 min read
So proud that 3(!) @StanfordAIMI publications are nominated for @AuntMinnie Scientific Paper of the Year!

@mattlungrenMD @josephpaulcohen @Dr_ASChaudhari @StanfordRad

All nominees are here: auntminnie.com/index.aspx?sec…

A thread with links to each paper: 1) Deep learning algorithms and are trained on data that lack diversity and come mostly from just 3 U.S. states: CA, MA, NY. @dr_amit_kaushal @Rbaltman @stanfordAIMI @StanfordRad

jamanetwork.com/journals/jama/…
Nov 2, 2019 4 tweets 1 min read
Most of what we do has no proven downstream outcome effects. IMHO the main reasons the radiology AI companies have not met investor expectations are:

1. high quality clinical training data is a lot harder to come by than radiology AI startups anticipated. 2. Commercial radiology AI algorithms mostly solve easier problems (bone age, brain hemorrhage) or use conveniently available data (lung nodules). They don't yet meet important radiology needs (work flow). So they are a "nice to have" feature, like auto-sync with priors in PACS.
Jun 27, 2019 12 tweets 5 min read
The @NYTimes article today describing the data sharing lawsuit against @UChicago has generated much discussion and also some confusion about patient data sharing for research. I thought it would be worth clarifying a few key issues 1/
nytimes.com/2019/06/26/tec… I have no special knowledge of the @Google and @UChicago deal, but data sharing is typically covered by a data use agreement that restricts use of shared data for a specific project, by specific people, and for a limited time. For example: ico.sites.stanford.edu/sites/g/files/… (Word doc) /2