For coming Monday's CDSl webinar, we'll be hosting Dr. Yun Liu from Google Health. Abstract: This talk will briefly cover two categories of our work: deep learning to identify dermatology conditions from
For coming Monday’s CDSl webinar, we’ll be hosting Dr. Yun Liu from Google Health.
This talk will briefly cover two categories of our work: deep learning to identify dermatology conditions from clinical images, and cancer prognostication from histopathology images. For the first talk, key background is that skin conditions are highly prevalent, however most cases are seen by general practitioners with lower diagnostic accuracy than dermatologists. We present a deep learning system (DLS) that distinguishes between 26 common skin conditions, while also providing a secondary prediction covering 419 skin conditions. On 963 validation cases, where a rotating panel of three board-certified dermatologists defined the reference standard, the DLS was non-inferior to six other dermatologists and superior to six primary care physicians (PCPs) and six nurse practitioners (NPs), highlighting the potential of the DLS to assist general practitioners in diagnosing skin conditions.
For the second work, we worked on predicting cancer prognosis from digitized images of histopathology samples. Our approach resolves around weakly-supervised approaches where the model is only provided information about survival outcomes without additional tissue-level annotations. We first prototyped our approach on TCGA across 10 cancer types, finding that the DLS was a significant predictor of survival in 5 of 10 cancer types, after adjusting for cancer type, stage, age, and sex. In followup work, we replicated our main findings with a larger cohort of intermediate-risk (stage II/III) colorectal cancer patients, and with full clinical cases instead of representative slides per case. We additionally showcased a generalizable method that identified a human-interpretable feature. This feature, “tumor-adipose feature”, was independently associated with survival, and reproducibly identified by both pathologists and non-pathologists, indicating promise in discovering novel, human-recognizable histoprognostic features for future research.
Yun is a staff research scientist in Google Health. In this role he focuses on developing and validating machine learning for medical imaging across multiple fields: pathology, ophthalmology, radiology, and dermatology. Yun completed his PhD at Harvard-MIT Health Sciences and Technology, where he worked on predictive risk modeling using biomedical signals, medical text, and billing codes. He has previously also worked on predictive modeling for nucleic acid sequences and protein structures. Yun completed a B.S. in Molecular and Cellular Biology and Computer Science at Johns Hopkins University.
Join Zoom Meeting
Meeting ID: 979 4193 1766
One tap mobile
+13017158592,,97941931766# US (Washington DC)
+13126266799,,97941931766# US (Chicago)
Thanks and have a good weekend,
(Monday) 3:00 pm - 4:00 pm
CDSLNCI CCR Cancer Data Science LabArati Rajeevan, email@example.com