Attention-based Generative Latent Replay: A Continual Learning Approach for WSI Analysis

TL;DR

The paper tackles domain shifts in WSI classification by introducing AGLR-CL, a buffer-free continual learning framework that uses MIL and Gaussian Mixture Models to model past data distributions. It employs attention-based filtering to select salient patches and generates synthetic past embeddings via a Generative Latent Replay mechanism, allowing replay without storing raw WSIs. Evaluations on MSI, TMB, and molecular-status prediction across TCGA, CPTAC, PAIP, and BCNB datasets demonstrate competitive performance with buffer-based methods and clear privacy advantages in domain adaptation. The approach advances privacy-preserving continual learning in computational pathology by enabling robust domain generalization without data retention. Overall, AGLR-CL represents a practical framework for continual learning in WSI analysis under privacy constraints.

Abstract

Whole slide image (WSI) classification has emerged as a powerful tool in computational pathology, but remains constrained by domain shifts, e.g., due to different organs, diseases, or institution-specific variations. To address this challenge, we propose an Attention-based Generative Latent Replay Continual Learning framework (AGLR-CL), in a multiple instance learning (MIL) setup for domain incremental WSI classification. Our method employs Gaussian Mixture Models (GMMs) to synthesize WSI representations and patch count distributions, preserving knowledge of past domains without explicitly storing original data. A novel attention-based filtering step focuses on the most salient patch embeddings, ensuring high-quality synthetic samples. This privacy-aware strategy obviates the need for replay buffers and outperforms other buffer-free counterparts while matching the performance of buffer-based solutions. We validate AGLR-CL on clinically relevant biomarker detection and molecular status prediction across multiple public datasets with diverse centers, organs, and patient cohorts. Experimental results confirm its ability to retain prior knowledge and adapt to new domains, offering an effective, privacy-preserving avenue for domain incremental continual learning in WSI classification.

Figures (2)

• Figure 1: Privacy-aware domain incremental AGLR-CL framework. Following WSI tessellation, a frozen FM-based encoder generates a sequence of tractable embeddings. A MIL aggregator is then trained on the current episode, with high-attention features selected to fit GMMs for patch embeddings and counts. In subsequent episodes, historical data is revisited by re-sampling synthetic WSIs using the per-episode GMMs.
• Figure 2: Attention heatmaps for AGLR-CL across domain shifts in MSI prediction. Attentions scores for WSIs from $\mathcal{D}_1$ (PAIP-CRC) by the model trained over five CL stages in a2 reflect that past knowledge acquired from $\mathcal{D}_1$ is preserved.