Deep Learning for Whole-Slide Tissue Histopathology Classification: A Comparative Study in the Identification of Dysplastic and Non-Dysplastic Barrett's Esophagus.

Rasoul Sali,Donald E Brown,Lubaina Ehsan,Shan Guleria,Tilak U Shah,Nazanin Moradinasab,Philip Fernandes,Sana Syed

doi:10.3390/jpm10040141

Abstract

The gold standard of histopathology for the diagnosis of Barrett’s esophagus (BE) is hindered by inter-observer variability among gastrointestinal pathologists. Deep learning-based approaches have shown promising results in the analysis of whole-slide tissue histopathology images (WSIs). We performed a comparative study to elucidate the characteristics and behaviors of different deep learning-based feature representation approaches for the WSI-based diagnosis of diseased esophageal architectures, namely, dysplastic and non-dysplastic BE. The results showed that if appropriate settings are chosen, the unsupervised feature representation approach is capable of extracting more relevant image features from WSIs to classify and locate the precursors of esophageal cancer compared to weakly supervised and fully supervised approaches.

Highlights

Barrett’s esophagus (BE) is a precancerous condition that results from damage to the lining of the squamous esophageal mucosa
The unsupervised approach with Gaussian mixture model (GMM) clustering demonstrated a considerable performance in the extraction of image features, which contributed to high separability between different classes. k-means was an exception—its graph does not show acceptable separability performance, at least between two hard-to-separate classes: dysplastic and non-dysplastic BE
We performed a comparative study on different feature representation approaches, including unsupervised, weakly supervised, and fully supervised approaches to classify precursors to esophageal cancer using whole-slide histopathology images

Summary

Introduction

Barrett’s esophagus (BE) is a precancerous condition that results from damage to the lining of the squamous esophageal mucosa. In order to increase sensitivity for dysplasia, guidelines recommend the Seattle protocol, which involves taking four-quadrant random biopsies at 1–2 cm intervals [2]. This protocol does not permit real-time diagnosis or therapy and is labor-intensive, leading to low adherence [3,4]. There is a major need to develop innovative computational methods to translate heterogeneous histopathological images into accurate and precise diagnostics. The development of such a methodology in high-dimensional clinical research will support precision medicine, with improved diagnostics, predictions, treatments, and patient clinical outcomes. The success of these approaches relies on the appropriateness of extracted morphological features for characterizing the images

Methods

Results

Discussion

Conclusion