Image Synthesis as a Pretext for Unsupervised Histopathological Diagnosis

TL;DR

This paper explores using high-resolution image synthesis with advanced generative models to improve unsupervised anomaly detection in digital pathology, demonstrating significant quality improvements over existing methods.

Contribution

It adapts and evaluates state-of-the-art generative models from face synthesis for digital pathology, enhancing normal appearance modeling and anomaly detection.

Findings

Multifold improvement in image synthesis quality and resolution

Superiority over current approaches in digital pathology

Validated against supervised models

Abstract

Anomaly detection in visual data refers to the problem of differentiating abnormal appearances from normal cases. Supervised approaches have been successfully applied to different domains, but require an abundance of labeled data. Due to the nature of how anomalies occur and their underlying generating processes, it is hard to characterize and label them. Recent advances in deep generative-based models have sparked interest in applying such methods for unsupervised anomaly detection and have shown promising results in medical and industrial inspection domains. In this work we evaluate a crucial part of the unsupervised visual anomaly detection pipeline, that is needed for normal appearance modeling, as well as the ability to reconstruct closest looking normal and tumor samples. We adapt and evaluate different high-resolution state-of-the-art generative models from the face synthesis domain and demonstrate their superiority over currently used approaches on a challenging domain of digital pathology. Multifold improvement in image synthesis is demonstrated in terms of the quality and resolution of the generated images, validated also against the supervised model.