Transfer Learning Strategies for Pathological Foundation Models: A Systematic Evaluation in Brain Tumor Classification

TL;DR

This study systematically evaluates transfer learning strategies for brain tumor classification using foundation models, revealing that simple methods like linear probing outperform fine-tuning, which often degrades performance, thus suggesting a shift in AI pathology approaches.

Contribution

It provides a comprehensive evaluation of transfer learning methods for brain tumor classification, highlighting the effectiveness of simple strategies over fine-tuning in clinical pathology.

Findings

Foundation models perform well with as few as 10 patches per case.

Linear probing is sufficient for effective classification.

Fine-tuning can degrade model performance.

Abstract

Foundation models pretrained on large-scale pathology datasets have shown promising results across various diagnostic tasks. Here, we present a systematic evaluation of transfer learning strategies for brain tumor classification using these models. We analyzed 254 cases comprising five major tumor types: glioblastoma, astrocytoma, oligodendroglioma, primary central nervous system lymphoma, and metastatic tumors. Comparing state-of-the-art foundation models with conventional approaches, we found that foundation models demonstrated robust classification performance with as few as 10 patches per case, despite the traditional assumption that extensive per-case image sampling is necessary. Furthermore, our evaluation revealed that simple transfer learning strategies like linear probing were sufficient, while fine-tuning often degraded model performance. These findings suggest a paradigm shift from "training encoders on extensive pathological data" to "querying pre-trained encoders with labeled datasets", providing practical implications for implementing AI-assisted diagnosis in clinical pathology.