Abstract:Domain shift in the field of histopathological imaging is a common phenomenon due to the intra- and inter-hospital variability of staining and digitization protocols. The implementation of robust models, capable of creating generalized domains, represents a need to be solved. In this work, a new domain adaptation method to deal with the variability between histopathological images from multiple centers is presented. In particular, our method adds a training constraint to the supervised contrastive learning approach to achieve domain adaptation and improve inter-class separability. Experiments performed on domain adaptation and classification of whole-slide images of six skin cancer subtypes from two centers demonstrate the method's usefulness. The results reflect superior performance compared to not using domain adaptation after feature extraction or staining normalization.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: in the field of histopathological images, the domain shift problem caused by differences in staining and digitization protocols between different hospitals. This domain shift makes it difficult for deep - learning models to generalize across multiple centers, thus affecting the accuracy and reliability of the models. Specifically, the paper aims to deal with the domain shift problem in multi - center whole - slide - image (WSI) classification by proposing a new domain adaptation method. This method is based on supervised contrastive learning and enhances the inter - class separability by introducing cross - domain constraints, thereby improving the generalization ability of the model on data from different centers. ### Main problem summary: 1. **Domain shift phenomenon**: Differences in staining and digitization protocols between different hospitals lead to distribution differences between training data and test data. 2. **Model generalization ability**: Existing deep - learning models experience a significant performance decline when faced with data from different centers and are unable to generalize effectively. 3. **Limitations of existing methods**: - Although the staining normalization method can reduce color differences, it will lose valuable information, resulting in a decline in classification performance. - The unsupervised contrastive learning method has high computational complexity and requires a large number of image samples, which is difficult to meet in histopathological images. ### Solutions proposed in the paper: - **Supervised Contrastive Domain Adaptation (SCDA)**: By introducing supervised contrastive learning and cross - domain constraints, the model can better handle the differences between different centers and improve the inter - class separability. - **Few - shot learning extension**: Extend the SCDA method to the few - shot learning paradigm to more efficiently adapt to data from new hospitals and reduce the dependence on large - scale data sets. ### Experimental verification: The paper conducted experiments on whole - slide - image data sets of skin cancer subtypes from two centers (Valencia and Granada) to verify the effectiveness of the proposed method. The results show that, compared with not using domain adaptation or only performing staining normalization, the SCDA method has a significant improvement in classification performance, especially in cross - hospital generalization. ### Formula representation: The supervised contrastive loss function is defined as: \[ L = \sum_{i \in I} L_i = \sum_{i \in I} -\log \frac{\exp(\frac{z_i \cdot z_p}{\tau})}{\sum_{a \in A(i)} \exp(\frac{z_i \cdot z_a}{\tau})} \] where: - \( z_i \) is the representation of sample \( i \), - \( z_p \) is the representation of another sample \( j\neq i \) belonging to the same class, - \( A(i) \) is the set of representations of all samples in the batch, - \( \tau \) is the temperature parameter used to scale the similarity between representations. By introducing cross - domain constraints, it is ensured that samples of the same class from different centers are clustered together in the feature space, thus effectively solving the domain shift problem.

Enhancing Whole Slide Image Classification through Supervised Contrastive Domain Adaptation

Adversarial Domain Adaptation for Classification of Prostate Histopathology Whole-Slide Images

Domain adaptation using optimal transport for invariant learning using histopathology datasets

A Closer Look at Domain Shift for Deep Learning in Histopathology

Self-Contrastive Weakly Supervised Learning Framework for Prognostic Prediction Using Whole Slide Images

Towards histopathological stain invariance by Unsupervised Domain Augmentation using generative adversarial networks

Enhancing Whole Slide Image Classification with Discriminative and Contrastive Learning

Contrastive image adaptation for acquisition shift reduction in medical imaging

The Whole Pathological Slide Classification via Weakly Supervised Learning

Color Deconvolution applied to Domain Adaptation in HER2 histopathological images

Multi-Cohort Framework with Cohort-Aware Attention and Adversarial Mutual-Information Minimization for Whole Slide Image Classification

Foundation Models for Slide-level Cancer Subtyping in Digital Pathology

Lesion-Aware Contrastive Representation Learning for Histopathology Whole Slide Images Analysis

Stain SAN: simultaneous augmentation and normalization for histopathology images

Learning generalizable AI models for multi-center histopathology image classification

Improving Stain Invariance of CNNs for Segmentation by Fusing Channel Attention and Domain-Adversarial Training

Graph Neural Networks for UnsupervisedDomain Adaptation of Histopathological ImageAnalytics

Mitigating the Influence of Domain Shift in Skin Lesion Classification: A Benchmark Study of Unsupervised Domain Adaptation Methods on Dermoscopic Images

Contrastive Centroid Supervision Alleviates Domain Shift in Medical Image Classification

Unsupervised domain adaptation for nuclei segmentation: Adapting from hematoxylin & eosin stained slides to immunohistochemistry stained slides using a curriculum approach

Adaptive Domain Generalization for Digital Pathology Images