MergeSurv: Merging-Based Continual Learning for Survival Analysis on Whole-Slide Images

2026-07-06

AI Summary

This work addresses the high computational cost and catastrophic forgetting associated with training separate survival analysis models for each cancer cohort on whole-slide images. To overcome these limitations, the study introduces— for the first time in computational pathology—a model merging mechanism within a continual learning framework based on a foundation pathological vision-language model. After independently fine-tuning on individual cohorts, model parameters are sequentially merged to construct a unified model without retaining historical data. The authors explore two inference strategies: One-for-All and Voting-Expert Aggregation. Experiments across four TCGA cohorts demonstrate that the proposed approach significantly outperforms naive fine-tuning as well as mainstream regularization- and replay-based continual learning methods, effectively mitigating catastrophic forgetting while maintaining computational efficiency and preserving data privacy.

Abstract

Survival analysis on Whole Slide Images (WSIs) is important in computational pathology for prognosis estimation and treatment planning. However, existing survival models are typically trained independently for each cancer cohort, making continual adaptation computationally expensive for gigapixel-scale WSIs. In this study, we propose MergeSurv, a merging-based continual learning framework for WSI survival analysis. A pathology vision-language foundation model is independently fine-tuned on each task, and the learned parameters are sequentially merged into a unified model without storing previous training data. We further investigate two inference strategies: One-for-All (OFA) and Voting-Expert Aggregation (VEA). Experiments on four TCGA cohorts demonstrate that MergeSurv outperforms naive fine-tuning as well as representative regularization-based and rehearsal-based continual learning methods, while effectively reducing catastrophic forgetting. The results suggest that model merging is a promising direction for scalable and privacy-preserving continual learning in computational pathology.

Problem

• survival analysis
• whole-slide images
• continual learning
• computational pathology
• catastrophic forgetting

Innovation

• model merging
• continual learning
• survival analysis
• whole-slide images
• vision-language foundation model