Workshop on Interpretability of Machine Intelligence in Medical Image Computing at MICCAI 2026

September 27th, 2026 / Strasbourg, France

iMIMIC
 

This workshop is part of the MICCAI 2026 conference.

Overview

Machine learning (ML) systems in medical imaging have evolved from task-specific networks to massive, general-purpose Foundation Models and Vision-Language Models (VLMs). While these systems achieve unprecedented performance, they introduce a new layer of complexity and opacity. The "black-box" nature of these billion-parameter models makes their behavior increasingly unpredictable, raising critical concerns about hallucinations, bias amplification, and robustness.

Developing methodologies for explaining model predictions is no longer just a "nice-to-have" feature for user trust; it is an imperative for AI safety and regulatory compliance. With frameworks like the EU AI Act classifying medical AI as high-risk, there is a legal mandate for traceability, transparency, and human oversight. Methodologies that allow physicians to validate model reasoning, identify failure cases, and quantify uncertainty are essential for the ethical deployment of these systems in clinical workflows.

Despite the urgency, the MICCAI community faces a gap between algorithmic performance and clinical interpretability. Standard post-hoc visualization techniques (e.g., saliency maps) are often insufficient for high-dimensional 3D/4D data, failing to provide the causal or concept-based insights required for informed medical decision-making.

The Workshop on Interpretability of Machine Intelligence in Medical Image Computing (iMIMIC) at MICCAI 2026 addresses these challenges by focusing on the next generation of XAI. We need to move beyond "where the model looked" to explaining "why the model decided." This includes inspecting if models align with pathophysiological domain knowledge, detecting shortcuts in training data, and handling the complexity of multimodal and longitudinal patient data. Ultimately, interpretability is the key to transforming raw predictive power into reliable, legally sound, and clinically actionable intelligence.

This workshop aims to foster discussion and presentation of ideas to tackle the many challenges and identify opportunities related to the interpretability of ML systems in the context of MICCAI. This marks the 9th edition of iMIMIC, and to our knowledge it remains the only forum at MICCAI dedicated exclusively to the interpretability and explainability of machine learning models.

The primary purposes of this workshop are:

  1. To introduce the unique challenges of interpreting Generative AI and Foundation Models in the context of MICCAI, distinguishing medical XAI from general computer vision.
  2. To move the state-of-the-art toward quantitative, causal, and mechanistic interpretability.
  3. To join researchers, clinicians, and regulatory experts to discuss the gap between technical explanations and human-centric clinical needs.
  4. To propose objective benchmarks and metrics for measuring the fidelity, robustness, and utility of explanations.

Covered topics include but are not limited to:

Room: Boston. All times are local. The iMIMIC program follows the joint poster session and coffee break.

Posters

All accepted papers are presented during the joint iMIMIC - UNSURE poster session and coffee break (15:30 - 16:00), and remain on display for the duration of the workshop. Presenters should mount their poster on the board matching the number below.

  1. Understanding Synergistic Interactions among Pathology Foundation Models via Adaptive Fusion
  2. Cellular-Communication-Level Interpretability for Pathology Foundation Models via Graph Distillation on Microenvironment
  3. A Probabilistic Source-Free Domain Adaptation Method for Concept Bottleneck Models
  4. Retrieving Patient-Specific Radiomic Feature Sets for Transparent Knee MRI Assessment
  5. RESHAPE: Representation Learning for the Explainability of Shapes
  6. Spatial Message Passing in Language Space for Pathology Image Interpretation
  7. Recursive Uncertainty-Gated Image Registration for Learning-Based Algorithms
  8. How Well Do Chest X-Ray VLM Attention Overlays Match Radiologist Boxes? A Cross-Model Audit and Radiologist Reader Study
  9. Less Annotation, More Interpretation: Prior-Guided Concept Bottleneck Models for Interpretable Cancer Imaging Diagnosis
  10. Faithful Faithfulness Evaluations: Challenges & Pitfalls Learned from a Breast MRI Case Study
  11. When Saliency Over-Credits Anatomy: Token-Level Evidence Flow in Frozen Medical Vision Transformers
  12. Towards Interpretable Foundation Models for Retinal Fundus Images
  13. Anatomical Information or Domain Shortcuts? Probing Image- and Shape-Based Embeddings in Ear CT with Landmark-Derived Features
  14. SAGE: Semantic Explainability of Attention-Based Survival Models in Computational Pathology
  15. P3CA: Encoder-Agnostic Interpretation of Vision Foundation Model Embeddings via Spatial Probing
  16. Attention Without Grounding: Causal Evaluation of Visual Explanations in Medical VLMs

Authors should prepare a manuscript of 8–10 pages, including references. The manuscript should be formatted and anonymized according to the Lecture Notes in Computer Science (LNCS) style. Proceedings will follow MICCAI Springer's publication model.

All submissions will be reviewed by three reviewers. Authors will be asked to disclose any potential conflicts of interest. The selection of papers will be based on their relevance to medical image analysis, the significance of the results, technical and experimental merit, and clear presentation.

Submit your manuscript via OpenReview: Submit to iMIMIC 2026.

(All times are 23:59 CEST)

The iMIMIC 2026 workshop will take place as part of the MICCAI 2026 conference, held 27 September – 1 October 2026 in Strasbourg, France.

More information regarding the venue can be found at the MICCAI 2026 conference website.

We gratefully acknowledge our Area Chairs and Program Committee members for their time, expertise, and thoughtful contributions to the iMIMIC 2026 review process.

Interested in participating and being a sponsor? Email us