mlsanityData to 7 Oct 2026

Beyond Attention Heatmaps: How to Get Better Explanations for Multiple Instance Learning Models in Histopathology

arXiv 2603.08328 ↗First posted 9 Mar 2026cs.CV cs.LGcited by 2

Not on the map. Why?

Abstract (arXiv)

Multiple instance learning (MIL) has enabled substantial progress in computational histopathology, where a large amount of patches from gigapixel whole slide images are aggregated into slide-level predictions. Heatmaps are widely used to validate MIL models and to discover tissue biomarkers. Yet, the validity of these heatmaps has barely been investigated. In this work, we introduce a general framework for evaluating the quality of MIL heatmaps without requiring additional labels. We conduct a large-scale benchmark experiment to assess six explanation methods across histopathology task types (classification, regression, survival), MIL model architectures (Attention-, Transformer-, Mamba-based), and patch encoder backbones (UNI2, Virchow2). Our results show that explanation quality mostly depends on MIL model architecture and task type, with perturbation ("Single"), layer-wise relevance propagation (LRP), and integrated gradients (IG) consistently outperforming attention-based and gradient-based saliency heatmaps, which often fail to reflect model decision mechanisms. We further demonstrate the advanced capabilities of the best-performing explanation methods: (i) We provide a proof-of-concept that MIL heatmaps of a bulk gene expression prediction model can be correlated with spatial transcriptomics for biological validation, and (ii) showcase the discovery of distinct model strategies for predicting human papillomavirus (HPV) infection from head and neck cancer slides. Our work highlights the importance of validating MIL heatmaps and establishes that improved explainability can enable more reliable model validation and yield biological insights, making a case for a broader adoption of explainable AI in digital pathology. Our code is provided in a public GitHub repository: https://github.com/bifold-pathomics/xMIL/tree/xmil-journal

Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Builds on 70 of its 146 references that are in the map. First posted on arXiv 9 Mar 2026.

Builds on

70 of 146 references.

References · 70

  1. 2017A Unified Approach to Interpreting Model Predictionscited by 6,488
  2. 2023Mamba: Linear-Time Sequence Modeling with Selective State Spacescited by 4,616
  3. 2017Axiomatic Attribution for Deep Networkscited by 3,350
  4. 2017Learning Important Features Through Propagating Activation Differencescited by 1,454
  5. 2018Sanity Checks for Saliency Mapscited by 1,149
  6. 2018Attention-based Deep Multiple Instance Learningcited by 1,077
  7. 2019Attention is not Explanationcited by 916
  8. 2020Quantifying Attention Flow in Transformerscited by 759
  9. 2020Data Efficient and Weakly Supervised Computational Pathology on Whole Slide Imagescited by 738
  10. 2019Attention is not not Explanationcited by 629
  11. 2021TransMIL: Transformer based Correlated Multiple Instance Learning for Whole Slide Image Classificationcited by 549
  12. 2020Captum: A unified and generic model interpretability library for PyTorchcited by 542
38 more
  1. 2020Dual-stream Multiple Instance Learning Network for Whole Slide Image Classification with Self-supervised Contrastive Learningcited by 473
  2. 2015Evaluating the visualization of what a Deep Neural Network has learnedcited by 416
  3. 2019Unmasking Clever Hans Predictors and Assessing What Machines Really Learncited by 359
  4. 2016DeepSurv: Personalized Treatment Recommender System Using A Cox Proportional Hazards Deep Neural Networkcited by 354
  5. 2017Methods for Interpreting and Understanding Deep Neural Networkscited by 342
  6. 2022From Anecdotal Evidence to Quantitative Evaluation Methods: A Systematic Review on Evaluating Explainable AIcited by 299
  7. 2022DTFD-MIL: Double-Tier Feature Distillation Multiple Instance Learning for Histopathology Whole Slide Image Classificationcited by 254
  8. 2022Quantus: An Explainable AI Toolkit for Responsible Evaluation of Neural Network Explanations and Beyondcited by 171
  9. 2024Virchow2: Scaling Self-Supervised Mixed Magnification Models in Pathologycited by 144
  10. 2020Higher-Order Explanations of Graph Neural Networks via Relevant Walkscited by 138
  11. 2023Visual Language Pretrained Multiple Instance Zero-Shot Transfer for Histopathology Imagescited by 132
  12. 2023CellViT: Vision Transformers for Precise Cell Segmentation and Classificationcited by 108
  13. 2024HEST-1k: A Dataset for Spatial Transcriptomics and Histology Image Analysiscited by 104
  14. 2022XAI for Transformers: Better Explanations through Conservative Propagationcited by 100
  15. 2023Modeling Dense Multimodal Interactions Between Biological Pathways and Histology for Survival Predictioncited by 92
  16. 2020Deep Learning-based Computational Pathology Predicts Origins for Cancers of Unknown Primarycited by 89
  17. 2024MambaMIL: Enhancing Long Sequence Modeling with Sequence Reordering in Computational Pathologycited by 87
  18. 2024Feature Re-Embedding: Towards Foundation Model-Level Performance in Computational Pathologycited by 66
  19. 2024Dynamic Graph Representation with Knowledge-aware Attention for Histopathology Whole Slide Image Analysiscited by 66
  20. 2021Cluster-to-Conquer: A Framework for End-to-End Multi-Instance Learning for Whole Slide Image Classificationcited by 65
  21. 2024A Clinical Benchmark of Public Self-Supervised Pathology Foundation Modelscited by 64
  22. 2022Additive MIL: Intrinsically Interpretable Multiple Instance Learning for Pathologycited by 58
  23. 2024Multimodal Prototyping for cancer survival predictioncited by 47
  24. 2023Structured State Space Models for Multiple Instance Learning in Digital Pathologycited by 41
  25. 2024RudolfV: A Foundation Model by Pathologists for Pathologistscited by 40
  26. 2020Building and Interpreting Deep Similarity Modelscited by 34
  27. 2022Guidelines and Evaluation of Clinical Explainable AI in Medical Image Analysiscited by 34
  28. 2025Distilling foundation models for robust and efficient models in digital pathologycited by 33
  29. 2019Resolving challenges in deep learning-based analyses of histopathological images using explanation methodscited by 32
  30. 2024MambaLRP: Explaining Selective State Space Sequence Modelscited by 31
  31. 2021Toward Explainable AI for Regression Modelscited by 30
  32. 2022ReMix: A General and Efficient Framework for Multiple Instance Learning based Whole Slide Image Classificationcited by 30
  33. 2024Decoupling Pixel Flipping and Occlusion Strategy for Consistent XAI Benchmarkscited by 28
  34. 2025Do Multiple Instance Learning Models Transfer?cited by 28
  35. 2021DeepSMILE: Contrastive self-supervised pre-training benefits MSI and HRD classification directly from H&E whole-slide images in colorectal and breast cancercited by 27
  36. 2022AdvMIL: Adversarial Multiple Instance Learning for the Survival Analysis on Whole-Slide Imagescited by 24
  37. 2024The Clever Hans Effect in Unsupervised Learningcited by 24
  38. 2020Predicting molecular phenotypes from histopathology images: a transcriptome-wide expression-morphology analysis in breast cancercited by 21

50 most cited of 70 All 70 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Symb-xMIL: Symbolic Explanations for Multiple Instance Learning in Digital Pathologycited by 0
  2. 2026In-Context Multiple Instance Learningcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Label-free Concept Based Multiple Instance Learning for Gigapixel Histopathology
  2. 2026SAGE: Semantic Explainability of Attention-Based Survival Models in Computational Pathology
  3. 2025Slide-Level Prompt Learning with Vision Language Models for Few-Shot Multiple Instance Learning in Histopathology
  4. 2026Mixture of Mini Experts: Overcoming the Linear Layer Bottleneck in Multiple Instance Learning
  5. 2024Aligning Knowledge Concepts to Whole Slide Images for Precise Histopathology Image Analysis
  6. 2026Multimodal LLMs Outperform Pathology Foundation Models in Cross-Domain Histological Similarity
  7. 2026A Generative Foundation Model for Multimodal Histopathology
  8. 2026Benchmarking Computational Pathology Foundation Models For Semantic Segmentation