Test-time Unsupervised Domain Adaptation

TL;DR

This paper proposes a test-time unsupervised domain adaptation framework for medical imaging, demonstrating that adapting models to individual target subjects improves performance over traditional methods that adapt to broader target domains.

Contribution

It introduces a novel test-time adaptation approach that focuses on individual subjects, showing improved results over conventional domain adaptation methods.

Findings

Test-time adaptation outperforms traditional domain adaptation for individual subjects.

Models adapted to specific subjects outperform those adapted to the entire target domain.

Test-time UDA is effective even with only a single target subject.

Abstract

Convolutional neural networks trained on publicly available medical imaging datasets (source domain) rarely generalise to different scanners or acquisition protocols (target domain). This motivates the active field of domain adaptation. While some approaches to the problem require labeled data from the target domain, others adopt an unsupervised approach to domain adaptation (UDA). Evaluating UDA methods consists of measuring the model's ability to generalise to unseen data in the target domain. In this work, we argue that this is not as useful as adapting to the test set directly. We therefore propose an evaluation framework where we perform test-time UDA on each subject separately. We show that models adapted to a specific target subject from the target domain outperform a domain adaptation method which has seen more data of the target domain but not this specific target subject. This result supports the thesis that unsupervised domain adaptation should be used at test-time, even if only using a single target-domain subject