High-throughput adaptive sampling for whole-slide histopathology image analysis (HASHI) via convolutional neural networks: Application to invasive breast cancer detection

Abstract

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 (200 × 200 pixels). Only recently, Fully convolutional networks (FCN) are able to deal with larger image sizes (500 × 500 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%.

Automated and Manual Quantification of Tumour Cellularity in Digital Slides for Tumour Burden Assessment

“…The advantages of this approach is that there is no hand-engineering of features involved, and instead appropriate image properties are captured in a model containing several layers. There has been previous work using deep neural networks in digital pathology [16,17] , and comparisons have shown we can achieve superior performance compared to traditional feature extraction methods [18][19][20]. In this study, we also found that by using deep neural networks, we could achieve strong agreements with scores produced by two study pathologists; achieving ICC agreements of 0.82, approaching the intra-rater agreement of 0.89 and with tighter upper and lower bounds, suggesting more stable measurements than can be achieved manually.…”

Prognostic Analysis of Histopathological Images Using Pre-Trained Convolutional Neural Networks

“…In order to avoid the information loss, WSIs are often divided into small patches (ex: 256 x 256 pixels) and each patch is analyzed individually as Region of interest (ROI). These ROIs can also be labeled using active learning [38] or by professional trained pathologists [39]. Then the integrated patch-level decision or object-level decision from averaging regions of patches representing WSIs are studied for the specific tasks [2].…”

Software tools for automated transmission electron dmicroscopy

“…This means that the main task in setting up an automated TEM workflow for a specific application is now reduced to developing a suitable image analysis routine. These routines are obviously not limited to the examples described in this article, but could also include advanced feature detection using machine learning 16,29 including convoluted neuronal networks 30,31 . These approaches will also help in reducing the need for manual identification of particles in cryo-EM by automatically picking holes or large particles.…”