RT Journal Article
SR Electronic
T1 H&E-stained Whole Slide Image Deep Learning Predicts SPOP Mutation State in Prostate Cancer
JF bioRxiv
FD Cold Spring Harbor Laboratory
SP 064279
DO 10.1101/064279
A1 Andrew J. Schaumberg
A1 Mark A. Rubin
A1 Thomas J. Fuchs
YR 2017
UL http://biorxiv.org/content/early/2017/03/03/064279.abstract
AB A quantitative model to genetically interpret the histology in whole microscopy slide images is desirable to guide downstream immuno- histochemistry, genomics, and precision medicine. We constructed a statistical model that predicts whether or not SPOP is mutated in prostate cancer, given only the digital whole slide after standard hematoxylin and eosin [H&E] staining. Using a TCGA cohort of 177 prostate cancer patients where 20 had mutant SPOP, we trained multiple ensembles of residual networks, accurately distinguishing SPOP mutant from SPOP wild type patients. We further validated our full metaensemble classifier on an independent test cohort from MSK-IMPACT of 152 patients where 19 had mutant SPOP. Mutants and non-mutants were accurately distinguished despite TCGA slides being frozen sections and MSK-IMPACT slides being formalin-fixed paraffin-embedded sections. Importantly, our method demonstrates tractable deep learning in this “small data” setting of 20 positive examples. To our knowledge, this is the first statistical model to predict a genetic mutation in cancer directly from the patient’s digitized H&E-stained whole microscope slide.Significance Statement The authors present the first automatic pipeline predicting gene mutation probability in cancer from digitized light microscopy slides having standard hematoxylin and eosin staining. To predict whether or not the speckle-type POZ protein [SPOP] gene is mutated in prostate cancer, the pipeline (i) identifies diagnostically salient regions in the whole slide at low magnification, (ii) identifies the salient region having the dominant tumor, (iii) within this region finds the high magnification subregion most enriched for abnormal cells, and (iv) trains ensembles of deep learning binary classifiers that together predict a confidence interval of mutation probability. Through deep learning on small datasets, this work enables fully-automated histologic diagnoses based on probabilities of underlying molecular aberrations. Such probabilities may directly guide immunohis- tochemistry choices, genetic tests, and precision medicine.