FunOTTA: On-the-Fly Adaptation on Cross-Domain Fundus Image via Stable Test-time Training.

2024-07-05

IEEE Transactions on Medical Imaging

AI Summary

To address performance degradation of pretrained models in cross-device and cross-institutional fundus image diagnosis caused by domain shift, this paper proposes an unsupervised test-time adaptation (TTA) method. The approach operates without target-domain labels and supports online adaptive updates. Key contributions include: (1) a novel dynamic memory bank ambiguity resolution mechanism to mitigate prior-knowledge bias; and (2) an incremental classifier adaptation objective grounded in reliable class-conditional probability estimation and consistency regularization. Evaluated on cross-domain benchmarks for two major ophthalmic diseases, the method significantly outperforms existing TTA approaches while maintaining compatibility with diverse backbone architectures. The source code is publicly available.

Technology Category

Machine Learning: Transfer, Domain Adaptation, Multi-Task LearningComputer Vision: Adversarial Attacks & RobustnessKnowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingUser Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendationSecurity and Privacy: Data transparency and provenance

Abstract

Fundus images are essential for the early screening and detection of eye diseases. While deep learning models using fundus images have significantly advanced the diagnosis of multiple eye diseases, variations in images from different imaging devices and locations (known as domain shifts) pose challenges for deploying pre-trained models in real-world applications. To address this, we propose a novel Fundus On-the-fly Test-Time Adaptation (FunOTTA) framework that effectively generalizes a fundus image diagnosis model to unseen environments, even under strong domain shifts. FunOTTA stands out for its stable adaptation process by performing dynamic disambiguation in the memory bank while minimizing harmful prior knowledge bias. We also introduce a new training objective during adaptation that enables the classifier to incrementally adapt to target patterns with reliable class conditional estimation and consistency regularization. We compare our method with several state-of-the-art test-time adaptation (TTA) pipelines. Experiments on cross-domain fundus image benchmarks across two diseases demonstrate the superiority of the overall framework and individual components under different backbone networks. Code is available at https://github.com/Casperqian/FunOTTA.

Problem

• Addressing domain shifts in fundus images from different devices and locations
• Enabling stable adaptation of diagnosis models to unseen environments
• Improving cross-domain generalization through test-time training techniques

Innovation

• Dynamic disambiguation in memory bank
• Minimizing harmful prior knowledge bias
• Reliable class conditional estimation with consistency