27 February 2018 Expert identification of visual primitives used by CNNs during mammogram classification
Author Affiliations +
This work interprets the internal representations of deep neural networks trained for classification of diseased tissue in 2D mammograms. We propose an expert-in-the-loop inter- pretation method to label the behavior of internal units in convolutional neural networks (CNNs). Expert radiologists identify that the visual patterns detected by the units are correlated with meaningful medical phenomena such as mass tissue and calcificated vessels. We demonstrate that several trained CNN models are able to produce explanatory descriptions to support the final classification decisions. We view this as an important first step toward interpreting the internal representations of medical classification CNNs and explaining their predictions.
© (2018) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Jimmy Wu, Jimmy Wu, Diondra Peck, Diondra Peck, Scott Hsieh, Scott Hsieh, Vandana Dialani, Vandana Dialani, Constance D. Lehman, Constance D. Lehman, Bolei Zhou, Bolei Zhou, Vasilis Syrgkanis, Vasilis Syrgkanis, Lester Mackey, Lester Mackey, Genevieve Patterson, Genevieve Patterson, "Expert identification of visual primitives used by CNNs during mammogram classification", Proc. SPIE 10575, Medical Imaging 2018: Computer-Aided Diagnosis, 105752T (27 February 2018); doi: 10.1117/12.2293890; https://doi.org/10.1117/12.2293890

Back to Top