Convolutional Neural Network - CS256-FC256Convolutional Neural Network (CNN) trained for patch-based classification of invasive breast cancer from histopathology digital images. The CNN architecture is 256 units in convolution and pooling layers, 256 units of fully connected layer and 2 units for output classification layer of softmax. The model was trained with Torch7.model_epoch25.netXML annotations by HG of WSIs from CINJXML region-based annotations by HG pathologist of whole-slide images (WSIs) from CINJ institution data cohort.XML_CINJ_HG.zipXML annotations by MF and NS of WSIs from CINJXML region-based annotations by MF and NS pathologists of whole-slide images (WSIs) from CINJ institution data cohort.XML_CINJ_MF+NS.zipXML annotations by HG of WSIs from TCGAXML region-based annotations by HG pathologist of whole-slide images (WSIs) from TCGA institution data cohort subset used in the paper.XML_TCGA_HG.zipTCGA scaled images195 TCGA scaled images used as D_test to test HASHI method.TCGA_imgs_idx5.zipUHCMC/CWRU scaled images110 UHCMC/CWRU scaled images used as D_2 dataset as part of Whole-Slide Image training data set.CWRU_imgs_idx8.zipCINJ scaled images40 CINJ scaled images used as D_4 dataset as part of Whole-slide Image validation data set.CINJ_imgs_idx5.zipHUP scaled images Part1120 of 239 HUP scaled images used as D_1 dataset as part of Whole-Slide Image training data set.HUP_imgs_idx5_Part1.zipHUP scaled images Part2119 of 239 HUP scaled images used as D_1 dataset as part of Whole-Slide Image training data set.HUP_imgs_idx5_Part2.zipHUP binary masks of annotationsHUP binary masks of manual annotations from pathologists.HUP_masks.zipUHCMC/CWRU binary masks of annotationsUHCMC/CWRU binary masks of manual annotations from pathologists.CWRU_masks.zipCINJ binary masks of annotationsCINJ binary masks of manual annotations from pathologists.CINJ_masks_HG.zipTCGA binary masks of annotationsTCGA binary masks of manual annotations from pathologists.TCGA_masks.zip

Precise detection of invasive cancer on whole-slide images (WSI) is a critical first step in digital pathology tasks of diagnosis and grading. Convolutional neural network (CNN) is the most popular representation learning method for computer vision tasks, which have been successfully applied in digital pathology, including tumor and mitosis detection. However, CNNs are typically only tenable with relatively small image sizes (200x200 pixels). Only recently, Fully convolutional networks (FCN) are able to deal with larger image sizes (500x500 pixels) for semantic segmentation. Hence, the direct application of CNNs to WSI is not computationally feasible because for a WSI, a CNN would require billions or trillions of parameters. To alleviate this issue, this paper presents a novel method, High-throughput Adaptive Sampling for whole-slide Histopathology Image analysis (HASHI), which involves: i) a new efficient adaptive sampling method based on probability gradient and quasi-Monte Carlo sampling, and, ii) a powerful representation learning classifier based on CNNs. We applied HASHI to automated detection of invasive breast cancer on WSI. HASHI was trained and validated using three different data cohorts involving near 500 cases and then independently tested on 195 studies from The Cancer Genome Atlas. The results show that (1) the adaptive sampling method is an effective strategy to deal with WSI without compromising prediction accuracy by obtaining comparative results of a dense sampling (~6 million of samples in 24 hours) with far fewer samples (~2,000 samples in 1 minute), and (2) on an independent test dataset, HASHI is effective and robust to data from multiple sites, scanners, and platforms, achieving an average Dice coefficient of 76%.