Evaluation of a Deep Neural Network for Automated Classification of Colorectal Polyps on Histopathologic Slides

Abstract

Are deep neural networks trained on data from a single institution for classification of colorectal polyps on digitized histopathology slides generalizable across multiple external institutions? Findings: A new deep neural network was developed based on 326 slide images from our institution to classify the four most common polyp types on digitized histopathology slides. In addition to evaluation on an internal test set of 157 slide images, we evaluated the model on an external test set of 238 slide images from 24 institutions across 13 states in the United States.This model achieved mean accuracies of 93.5% and 87.0% on the internal and external test sets, respectively, which were comparable with the performance of local pathologists on these test sets.Meaning: Deep neural networks could provide a generalizable approach for the classification of colorectal polyps on digitized histopathology slides and if confirmed in clinical trials, could potentially improve the efficiency, reproducibility, and accuracy of one of the most common cancer screening procedures.

Application of artificial intelligence in gastroenterology: Potential role in clinical practice

“…As the vast majority (> 80%) of the colorectal adenomas being detected at screening are actually small in size and without advanced histology, which means that these large numbers of specimens sent for pathological assessment are low‐risk lesions but occupy tremendous working hours of pathologists. In a recent study by Wei et al ., it was demonstrated that the deep neural network model achieved an accuracy of 87.0% (95% confidence interval, 82.7–91.3%) in classifying colorectal polyp using the digital histopathological slide images, which was comparable with local ‘pathologists’ accuracy of 86.6% (95% confidence interval, 82.3–90.9%) 29 . If such a model is validated by large‐scale prospective clinical trials and successfully integrated into the real‐world workflow of pathological reading, it may not only largely improve the efficiency but also preserve the constrained working hours of the pathologists for a detailed assessment of the high‐risk specimens.…”

Classification of colorectal tissue images from high throughput tissue microarrays by ensemble deep learning methods

“…In our experience, even on whole slide images, this is an extremely challenging issue, namely the differentiation between dysplastic glands found in adenomas/ precursor lesions and, especially, well-differentiated carcinomas. There have been publications investigating differences in adenoma type 38,39 , however the distinction between dysplastic lesions and cancers still remains a considerable challenge. To summarize, the novel aspects of this study are as follows:…”

Decidual Vasculopathy Identification in Whole Slide Images Using Multiresolution Hierarchical Convolutional Neural Networks

“…Because variations in scanning techniques have the possibility of causing biases between local institutions, future development would benefit from crossinstitutional data sources to help validate the generalizability of the algorithm. However, some recent research 42,43 has found that regularized convolutional neural networks trained on single-institution data were robust to cohort variations during validation on data from other institutions.…”