Predicting Metastatic Risk from Primary Tissue Architecture via Distance-Aware Spatial Modeling

2026-06-26

AI Summary

This study addresses the frequent neglect of spatial tissue architecture in existing methods for predicting distant metastasis risk from primary tumor histopathology images. To overcome this limitation, the authors propose DTMf-MIL, a novel multiple instance learning (MIL) framework that explicitly incorporates signed distance functions (SDFs) to model the geometric arrangement among tumor cells, cancer-associated fibroblasts, and infiltrating lymphocytes, thereby embedding spatial priors into feature learning. By moving beyond the conventional assumption of unordered bags, DTMf-MIL employs a distance-aware mechanism to capture structural characteristics at the interfaces of the tumor microenvironment. Experimental results demonstrate that DTMf-MIL consistently outperforms state-of-the-art methods that disregard spatial information across multiple public clinical benchmarks, confirming the substantial and consistent benefit of spatial awareness in metastasis risk prediction.

Abstract

Predicting the risk of distant metastasis from primary tumor tissue histology is a critical yet challenging task in computational pathology. Multiple Instance Learning (MIL) approaches can attend to subdomains in tumor regions that harbor features of metastatic cancer progression. However MIL models treat tissue patches as unordered bags, discarding the spatial layout that defines the metastatic potential. We propose that metastatic risk is inherently dictated by the geometric arrangement of the tumor microenvironment at the interface with tumor cells. Our model is designed to explicitly capture the spatial relationships between tumor cells, tumor associated fibroblasts and infiltrating lymphocytes. For this purpose, we propose Distance aware Tissue Modeling for Multiple Instance Learning(DTMf-MIL), a novel method that reinforces visual features with explicit spatial priors. By computing signed distance functions (SDF) relative to tissue phenotypes, our model learns to recognize structural signatures of metastatic risk. This geometric awareness translates directly to superior clinical performance as DTMf-MIL significantly outperforms state-of-the-art methods that ignore spatial layout on metastasis prediction from tissue in the primary tumor. We further validate our approach on public benchmarks, demonstrating that spatial awareness consistently improves diagnostic accuracy across diverse clinical tasks.

Problem

• metastatic risk prediction
• spatial modeling
• computational pathology
• tumor microenvironment
• histology

Innovation

• spatial modeling
• signed distance function
• multiple instance learning
• tumor microenvironment
• metastasis prediction