Focus on Focus: Focus-oriented Representation Learning and Multi-view Cross-modal Alignment for Glioma Grading

TL;DR

The paper introduces Focus on Focus (FoF), a framework for glioma grading that combines Focus-oriented Representation Learning (FRL) and Multi-view Cross-modal Alignment (MCA) to strengthen histopathology representations and directly align them with molecular biomarkers. FRL identifies diagnostic regions via a pixel-wise contribution map and enforces consistency across global, positive, and negative regions, while MCA uses biomarker-aware projections and supervised contrastive learning to bind histology to genomic statuses such as IDH, 1p/19q, and CNVs. Trained end-to-end on paired pathology-genomic data but capable of pathology-only inference, FoF achieves state-of-the-art performance on the TCGA-GBM-LGG dataset, outperforming both histopathology-only baselines and existing multimodal methods. The approach demonstrates notable improvements in accuracy, AUC, AP, and Kappa, and yields interpretable visualizations that highlight key diagnostic regions, indicating strong clinical relevance for resource-limited settings where molecular biomarkers may be unavailable.

Abstract

Recently, multimodal deep learning, which integrates histopathology slides and molecular biomarkers, has achieved a promising performance in glioma grading. Despite great progress, due to the intra-modality complexity and inter-modality heterogeneity, existing studies suffer from inadequate histopathology representation learning and inefficient molecular-pathology knowledge alignment. These two issues hinder existing methods to precisely interpret diagnostic molecular-pathology features, thereby limiting their grading performance. Moreover, the real-world applicability of existing multimodal approaches is significantly restricted as molecular biomarkers are not always available during clinical deployment. To address these problems, we introduce a novel Focus on Focus (FoF) framework with paired pathology-genomic training and applicable pathology-only inference, enhancing molecular-pathology representation effectively. Specifically, we propose a Focus-oriented Representation Learning (FRL) module to encourage the model to identify regions positively or negatively related to glioma grading and guide it to focus on the diagnostic areas with a consistency constraint. To effectively link the molecular biomarkers to morphological features, we propose a Multi-view Cross-modal Alignment (MCA) module that projects histopathology representations into molecular subspaces, aligning morphological features with corresponding molecular biomarker status by supervised contrastive learning. Experiments on the TCGA GBM-LGG dataset demonstrate that our FoF framework significantly improves the glioma grading. Remarkably, our FoF achieves superior performance using only histopathology slides compared to existing multimodal methods. The source code is available at https://github.com/peterlipan/FoF.

Figures (4)

• Figure 1: The FoF framework. FRL identifies the positive $\boldsymbol{x}_{\rm{pos}}$ and negative regions $\boldsymbol{x}_{\rm{neg}}$, encouraging the model to focus on the most important regions with a consistency constrict. The MCA module correlates the histopathology representations $\{\boldsymbol{v}_{\rm{glb}}, \boldsymbol{v}_{\rm{pos}}, \boldsymbol{v}_{\rm{neg}} \}$ with molecular biomarkers.
• Figure 2: Visualization of (a) the input pathology slide, (b) the CAM produced by baseline method ViT-Tiny, and (c) the CAM generated by the proposed FoF framework. As illustrated, FoF focuses on microvascular proliferation, which leads to the diagnosis of Glioblastoma (Grade IV).
• Figure 3: Comparisons with the baseline on the molecular-pathology knowledge alignment.
• Figure 4: Comparison with the baseline on model focus. As illustrated, our FoF framework concentrates more precisely on the key structures of Glioblastoma, such as microvascular proliferation and pseudopalisadeing necrosis.