Advancing Cross-Organ Domain Generalization with Test-Time Style Transfer and Diversity Enhancement

2025-03-24

AI Summary

To address degraded cross-organ domain generalization in computational pathology, this paper proposes a test-time domain adaptation framework that operates without target-domain labels. Our method dynamically aligns source and target features in a shared latent space via Test-Time Bidirectional Style Transfer (T3s), the first such approach for pathology. To enhance discriminability and diversity of style representations, we introduce the Cross-domain Style Orthogonal Diversification Module (CSDM), which jointly enforces orthogonality constraints on style bases and incorporates Low-Rank Adaptation (LoRA). Additionally, multi-scale data augmentation is integrated to improve robustness. Evaluated on three entirely unseen cross-organ pathological datasets, our framework achieves average accuracy improvements of 5.2–8.7% over current state-of-the-art methods, demonstrating superior generalization under domain shift.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Multi-modal Vision
• Application Domains: Humanities & Computational Social Science

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Responsible Web: Algorithmic accountability and transparency on the web
• Security and Privacy: Data transparency and provenance

Abstract

Deep learning has made significant progress in addressing challenges in various fields including computational pathology (CPath). However, due to the complexity of the domain shift problem, the performance of existing models will degrade, especially when it comes to multi-domain or cross-domain tasks. In this paper, we propose a Test-time style transfer (T3s) that uses a bidirectional mapping mechanism to project the features of the source and target domains into a unified feature space, enhancing the generalization ability of the model. To further increase the style expression space, we introduce a Cross-domain style diversification module (CSDM) to ensure the orthogonality between style bases. In addition, data augmentation and low-rank adaptation techniques are used to improve feature alignment and sensitivity, enabling the model to adapt to multi-domain inputs effectively. Our method has demonstrated effectiveness on three unseen datasets.

Problem

• Addressing domain shift in cross-organ computational pathology tasks
• Enhancing model generalization with test-time style transfer
• Improving multi-domain adaptation via style diversification and augmentation

Innovation

• Test-time style transfer with bidirectional mapping
• Cross-domain style diversification for orthogonality
• Data augmentation and low-rank adaptation techniques