Description
The inter-observer disagreement on lung cancer histopathology diagnosis is recognized and the morphological patterns associated with the molecular subtypes have not been studied. In this study, we built convolutional neural network models to distinguish tumor from adjacent dense benign regions (AUC > 0.935), recapitulated diagnosis (AUC > 0.88), with the results validated in an independent cohort (AUC > 0.85). We further demonstrated that quantitative histopathology features identified the major transcriptomic subtypes of lung cancer (P<0.01).
Learning Objective: After participating in this session, the learner should be better able to:
* Recognize the utility of convolutional neural network in correlating microscopic tissue morphology with the types and subtypes of non-small cell lung cancer.
* Understand the difference between convolutional neural networks and conventional machine learning methods for image analyses.
* Appreciate the associations between microscopic morphology and gene expression subtypes in non-small cell lung cancer.
Authors:
Kun-Hsing Yu (Presenter)
Harvard Medical School
Feiran Wang, Stanford University
Gerald Berry, Stanford University
Christopher Ré, Stanford University
Russ Altman, Stanford University
Michael Snyder, Stanford University
Isaac Kohane, Harvard Medical School
Presentation Materials: