MorphDistill: Distilling Unified Morphological Knowledge from Pathology Foundation Models for Colorectal Cancer Survival Prediction

2026-04-07

AI Summary

Existing foundation models in computational pathology often overlook organ-specific morphological features, limiting their prognostic performance in colorectal cancer survival prediction. To address this, this work proposes MorphDistill, a two-stage framework that first distills complementary knowledge from multiple foundation models into a compact, colorectal cancer–specialized encoder via a novel dimension-agnostic multi-teacher relational distillation strategy. In the second stage, it aggregates multi-instance features at the whole-slide image level through supervised contrastive regularization and attention mechanisms to predict five-year survival outcomes. This approach preserves inter-sample relationships without explicit feature alignment, enabling task-oriented unified representation learning. Evaluated on the Alliance/CALGB cohort, the method achieves an AUC of 0.68 (an 8% relative improvement) and a C-index of 0.661; external validation on TCGA yields a C-index of 0.628, significantly outperforming all baseline models.

Abstract

Background: Colorectal cancer (CRC) remains a leading cause of cancer-related mortality worldwide. Accurate survival prediction is essential for treatment stratification, yet existing pathology foundation models often overlook organ-specific features critical for CRC prognostication. Methods: We propose MorphDistill, a two-stage framework that distills complementary knowledge from multiple pathology foundation models into a compact CRC-specific encoder. In Stage I, a student encoder is trained using dimension-agnostic multi-teacher relational distillation with supervised contrastive regularization on large-scale colorectal datasets. This preserves inter-sample relationships from ten foundation models without explicit feature alignment. In Stage II, the encoder extracts patch-level features from whole-slide images, which are aggregated via attention-based multiple instance learning to predict five-year survival. Results: On the Alliance/CALGB 89803 cohort (n=424, stage III CRC), MorphDistill achieves an AUC of 0.68 (SD 0.08), an approximately 8% relative improvement over the strongest baseline (AUC 0.63). It also attains a C-index of 0.661 and a hazard ratio of 2.52 (95% CI: 1.73-3.65), outperforming all baselines. On an external TCGA cohort (n=562), it achieves a C-index of 0.628, demonstrating strong generalization across datasets and robustness across clinical subgroups. Conclusion: MorphDistill enables task-specific representation learning by integrating knowledge from multiple foundation models into a unified encoder. This approach provides an efficient strategy for prognostic modeling in computational pathology, with potential for broader oncology applications. Further validation across additional cohorts and disease stages is warranted.

Problem

• colorectal cancer
• survival prediction
• pathology foundation models
• morphological knowledge
• prognostication

Innovation

• knowledge distillation
• foundation models
• morphological representation
• computational pathology
• survival prediction