Gradient Alignment Improves Test-Time Adaptation for Medical Image Segmentation

TL;DR

This paper introduces GraTa, a novel test-time adaptation method for medical image segmentation that uses gradient alignment and a dynamic learning rate to better handle domain shifts, improving model performance during inference.

Contribution

The paper proposes a gradient alignment-based TTA method with a dynamic learning rate, enhancing optimization and domain adaptation in medical image segmentation.

Findings

GraTa outperforms state-of-the-art TTA methods on benchmark datasets.

Gradient alignment improves the direction of model updates during test-time.

Dynamic learning rate based on cosine similarity enhances adaptation efficiency.

Abstract

Although recent years have witnessed significant advancements in medical image segmentation, the pervasive issue of domain shift among medical images from diverse centres hinders the effective deployment of pre-trained models. Many Test-time Adaptation (TTA) methods have been proposed to address this issue by fine-tuning pre-trained models with test data during inference. These methods, however, often suffer from less-satisfactory optimization due to suboptimal optimization direction (dictated by the gradient) and fixed step-size (predicated on the learning rate). In this paper, we propose the Gradient alignment-based Test-time adaptation (GraTa) method to improve both the gradient direction and learning rate in the optimization procedure. Unlike conventional TTA methods, which primarily optimize the pseudo gradient derived from a self-supervised objective, our method incorporates an auxiliary gradient with the pseudo one to facilitate gradient alignment. Such gradient alignment enables the model to excavate the similarities between different gradients and correct the gradient direction to approximate the empirical gradient related to the current segmentation task. Additionally, we design a dynamic learning rate based on the cosine similarity between the pseudo and auxiliary gradients, thereby empowering the adaptive fine-tuning of pre-trained models on diverse test data. Extensive experiments establish the effectiveness of the proposed gradient alignment and dynamic learning rate and substantiate the superiority of our GraTa method over other state-of-the-art TTA methods on a benchmark medical image segmentation task. The code and weights of pre-trained source models are available at https://github.com/Chen-Ziyang/GraTa.