Test-time image-to-image translation ensembling improves out-of-distribution generalization in histopathology

TL;DR

This paper introduces a test-time image-to-image translation ensembling method that improves out-of-distribution generalization in histopathology by projecting images into source domains before classification.

Contribution

It proposes a novel test-time data augmentation technique using multi-domain image translation to enhance robustness across unseen protocols in histopathology.

Findings

Significant performance boost in domain generalization tasks.

Outperforms conventional domain generalization and augmentation methods.

Effective across multiple histopathology tasks.

Abstract

Histopathology whole slide images (WSIs) can reveal significant inter-hospital variability such as illumination, color or optical artifacts. These variations, caused by the use of different scanning protocols across medical centers (staining, scanner), can strongly harm algorithms generalization on unseen protocols. This motivates development of new methods to limit such drop of performances. In this paper, to enhance robustness on unseen target protocols, we propose a new test-time data augmentation based on multi domain image-to-image translation. It allows to project images from unseen protocol into each source domain before classifying them and ensembling the predictions. This test-time augmentation method results in a significant boost of performances for domain generalization. To demonstrate its effectiveness, our method has been evaluated on 2 different histopathology tasks where it outperforms conventional domain generalization, standard H&E specific color augmentation/normalization and standard test-time augmentation techniques. Our code is publicly available at https://gitlab.com/vitadx/articles/test-time-i2i-translation-ensembling.