Automated Localization of Breast Ductal Carcinoma in Situ in Whole Slide Images
Nikhil Seth · TSpace (University of Toronto) · 2019
Duct segmentation in whole slide images (WSIs) is an important step needed to analyze breast ductal carcinoma in-situ (DCIS), an early form of breast cancer. Here, we trained several U-Net architectures – deep convolutional neural networks designed to output probability maps – to segment DCIS in WSIs and validate the optimal patch field of view necessary to achieve superior accuracy at the slide-level. A U-Net trained at 5x achieved the best test results (DSC = 0.771, F1 = 0.601), implying the U-Net benefits from seeing wider contextual information. A custom U-Net based architecture, trained to incorporate patches from all available resolutions, achieved test results of DSC = 0.759 (F1 = 0.682), showing improvement in the model’s duct detecting capabilities. Both architectures showed comparable performance to a second expert annotator on an independent test set. This is preliminary work for a pipeline targeted at predicting recurrence risk in DCIS patients.