# SAGE: Semantic Explainability of Attention-Based Survival Models in Computational Pathology

arXiv: https://arxiv.org/abs/2608.02803 · HTML: https://mlsanity.com/paper/2608.02803/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 9 of its 18 references that are in the map.
- First posted on arXiv 3 Aug 2026.

## Builds on (most cited first, 9 of 9)

- [Adam: A Method for Stochastic Optimization](https://mlsanity.com/paper/1412.6980/) (2014)
- [Concept Bottleneck Models](https://mlsanity.com/paper/2007.04612/) (2020)
- [HEST-1k: A Dataset for Spatial Transcriptomics and Histology Image Analysis](https://mlsanity.com/paper/2406.16192/) (2024)
- [Interpretable Vision-Language Survival Analysis with Ordinal Inductive Bias for Computational Pathology](https://mlsanity.com/paper/2409.09369/) (2024)
- [Aligning Knowledge Concepts to Whole Slide Images for Precise Histopathology Image Analysis](https://mlsanity.com/paper/2411.18101/) (2024)
- [Label-free Concept Based Multiple Instance Learning for Gigapixel Histopathology](https://mlsanity.com/paper/2501.02922/) (2025)
- [Leveraging Foundation Models for Histological Grading in Cutaneous Squamous Cell Carcinoma using PathFMTools](https://mlsanity.com/paper/2511.19751/) (2025)
- [VLEER: Vision and Language Embeddings for Explainable Whole Slide Image Representation](https://mlsanity.com/paper/2502.20850/) (2025)
- [Unifying Multiple Foundation Models for Advanced Computational Pathology](https://mlsanity.com/paper/2503.00736/) (2025)

## Abstract (arXiv)

Attention-based multiple instance learning (ABMIL) is the predominant approach for slide-level prediction in computational pathology, yet its attention maps provide only local explanations: they indicate where a model focuses but not which histological features drive its predictions or how the model behaves across a patient cohort. We present Semantic Attention Global Explanations (SAGE), a post-hoc framework that extracts global, language-grounded explanations from a frozen ABMIL model. Using a pathology vision-language model, SAGE scores image patches against a dictionary of 25 histological concepts, aggregates these scores according to the model's learned attention, and quantifies how each concept relates to prediction risk across a cohort. Applied to survival prediction using seven TCGA cancer cohorts and three foundation models, SAGE recovered established prognostic features, such as the adverse association of necrosis, while revealing cancer-specific biology, including a favorable angiogenic signature in renal cell carcinoma consistent with known molecular subtypes. Ablation studies demonstrated that these associations depend on the model's learned attention rather than concept prevalence alone, and that the concept dictionary captures much of the prognostic information encoded by the foundation model features. Through semantically-grounded explanations, SAGE provides a scalable, model-agnostic framework for understanding what ABMIL survival models learn, enabling pathologists to interpret model behavior at the cohort level and offering the potential for biomarker identification.