Special Sessions

Abstract: This session will highlight funding opportunities at the National Institute of Biomedical Imaging and Bioengineering (NIBIB) and provide valuable insights for both new and established investigators aiming to apply to funding. Experts from NIBIB will share strategies for navigating the NIH grant system, preparing competitive applications, and introduce their programs and funding opportunities. Join this interactive session to enhance your grant-writing skills and discover resources to support your pursuit of research funding. The session will also highlight the new Precision Medicine with AI: Integrating Imaging with Multimodal Data (PRIMED-AI) initiative, as well as other trans-NIH funding opportunities. The session will include a Q&A segment where you can ask questions and interact with the panel members.

Thursday | April 9, 2026 | 15:00 – 16:00

Abstract: Foundation models have gained substantial interest in the medical imaging field. While the potential opportunities are substantial, foundation models require large amounts of data which is a challenge that needs to be addressed for proper development. This special session explores the technical, methodological, and ethical challenges of preparing large-scale medical imaging datasets for the development and validation of foundation models. Topics of interest include strategies for multi-institutional data integration, privacy-preserving de-identification, and scalable data search and curation pipelines. The session also seeks to address how dataset design influences model generalizability, fairness, and reproducibility, as well as the governance frameworks needed to ensure responsible use of imaging data. The goal is that participants will gain practical insights and understand how to shape emerging best practices for large-scale medical AI development.

• Organizers: Martin J. Willemink (Segmed)
• Session chairs: Martin J. Willemink (Segmed)

Keynotes information:

• Title 1: Setting the stage: why data preparation is the bottleneck for medical imaging foundation models
• Description 1: This talk will elaborate on best practices, challenges, and innovations in preparing medical data for foundation model development, from raw clinical datasets to curated, high-quality inputs
• Speaker 1: Martin J. Willemink (Segmed)

Full session schedule:

• 15:00 – Keynote 1
• 15:10 – #1571219244 – When Specialization Helps (and Hurts): Cross-Modality Transfer in Ophthalmic Imaging with Foundation Models
• 15:20 – #1571220007 – Synthetic Volumetric Data Generation Enables Zero-Shot Generalization of Foundation Models in 3D Medical Image Segmentation
• 15:30 – #1571220596 – SAFM-CTA: Segmentation-Augmented Foundation Model for CT-to-CTA Translation
• 15:40 – #1571222234 – Lost in Distortion: Uncovering the Domain Gap Between Computer Vision and Brain Imaging – a Study on Pretraining for Age Prediction
• 15:50 – #1571222527 – Hospital-Specific Bias in Patch-Based Pathology Models

Thursday | April 9, 2026 | 16:30 – 17:30

Abstract: The ISBI 2026 special session `Safety and Reliability of Medical Imaging Technology’ will explore challenges and solutions to building trustworthy AI systems for biomedical imaging. Ensuring the fairness, safety, and robustness of clinic-facing AI has become critical. We invite novel research contributions on topics such as bias and fairness, out-of-distribution and failure detection, uncertainty quantification, or the use of explainability in biomedical applications.

• Organizers: Irina Voiculescu (University of Oxford), Vito Paolo Pastore (University of Genoa)
• Session chairs: Irina Voiculescu (University of Oxford), Vito Paolo Pastore (University of Genoa)

Keynotes information:

• Title 1: Counterfactual Contrastive Analysis: Learning Common and Salient Generative Factors Between Two Image Sets
• Description 1: A model-agnostic framework for generating visual counterfactual explanations with contrastive analysis.
• Speaker 1: Pietro Gori (LTCI, Télécom Paris, Institut Polytechnique de Paris)
• Title 2: The Map Exists. Let's Build it: Closing the Gap Between Trustworthy AI Principles and Practice
• Description 2: Closing the gap between agreed principles and real-world implementation of trustworthy AI in medical imaging.
• Speaker 2: Maria A. Zuluaga (Data Science Department, EURECOM, Biot, France)

Full session schedule:

• 16:30 – Keynote 1
• 16:50 – Keynote 2
• 17:10 – #1571226517 – Penalty-Based Multi-Objective Optimization for Robust Data Splitting to Improve Consistency in Deep Learning Models
• 17:20 – #1571221618 – Investigating Label Bias and Representational Sources of Age-Related Disparities in Medical Segmentation

Friday | April 10, 2026 | 8:00 – 9:00

Abstract: This session will explore emerging methodologies for integrating multi-omics, imaging, and clinical data to build digital twins that can simulate individual disease trajectories and therapeutic responses. By convening experts in bioinformatics, radiomics, systems biology, and AI, the session aims to promote cross-disciplinary dialogue and identify key computational and translational challenges. Discussions will focus on strategies for data harmonization, model interpretability, and validation pipelines to ensure robustness, reproducibility, and clinical applicability of digital twin frameworks.

• Organizers: Valentina Giannini (University of Turin), Samanta Rosati (Politecnico di Torino) and Giulia Nicoletti (University of Turin)
• Session chairs: Valentina Giannini (University of Turin)

Keynotes information:

• Title 1: Precision Oncology Digital Twins: Integrating Multimodal Data for Next-Generation Care
• Description 1: An overview of methods for multimodal data aggregation in digital twin design, with a focus on enabling therapy de-escalation in breast cancer.
• Speaker 1: Valentina Giannini (University of Turin)

Full session schedule:

• 10:00 – Keynote 1
• 10:20 – #1571221287 – Hypernetwork-Based Adaptive Aggregation for Multimodal Multiple-Instance Learning in Predicting Coronary Calcium Debulking
• 10:30 – #1571221535 – Multi-Modal Domain Adaptation for Cancer Diagnosis via Integration of Pathological Images and Clinical Text Data
• 10:40 – #1571225579 – Multimodal Oncology Agent for IDH1 Mutation Prediction in Low-Grade Glioma
• 10:50 – #1571226630 – Adaptive Clinical-Aware Latent Diffusion for Multimodal Brain Image Generation and Missing Modality Imputation

Friday | April 10, 2026 | 10:30 – 11:30

Abstract: This special session brings together recent advances in personalized incremental (continual) learning and federated learning to address two persistent barriers in biomedical imaging: strict data privacy constraints and chronic data scarcity across institutions. We will showcase methods that enable AI models to adapt over time to longitudinal, non-IID, multi-center data without centralizing patient information, while remaining robust to scanner/protocol shifts. The session will feature contributions spanning algorithms, evaluation protocols, and real clinical use cases (CT/MRI, interventional imaging, oncology), highlighting pathways toward deployable, regulation-aware medical AI. Overall, the goal is to articulate a practical blueprint for privacy-preserving, data-efficient imaging AI in real-world healthcare settings.

• Organizers: Raffaele Mineo (University Campus Bio-Medico of Rome), Simone Palazzo (University of Catania), Giovanni Bellitto (University of Catania), Amelia Sorrenti (University Campus Bio-Medico of Rome), Federica Proietto Salanitri (University of Catania), Concetto Spampinato (University of Catania)
• Session chairs: Concetto Spampinato (University of Catania), Ulas Bagci (Northwestern University)

Keynotes information:

• Title 1: Trustworthy Continual Learning for Medical Image Analysis
• Description 1: A talk on building trustworthy medical imaging AI that can continuously adapt to shifting data, new tasks, and privacy constraints while remaining reliable, secure, and clinically effective.
• Speaker 1: Xiaofeng Liu (Yale University)

Full session schedule:

• 10:30 – Keynote 1
• 10:50 – #1571222209 – Continual Pretraining Prostate MRI ViT Encoder for Task Alignment
• 11:00 – #1571217774 – FedHarmo: Harmonizing Global Aggregation and Local Alignment in Cross-Domain Federated Medical Image Segmentation
• 11:10 – #1571220910 – Hyperbolic Model Aggregation for Federated Learning in fMRI
• 11:20 – #1571226519 – Federated Few-Shot Segmentation for Rare Brain Tumours via Weighted Aggregation and Prototype-Guided Adaptation

Abstract: This special session focuses on the growing problem of “data crimes” – scenarios where developers of artificial intelligence (AI) models naively use public datasets for tasks they were never designed to support, leading to biased outcomes and misleading algorithmic performance. Focusing on image reconstruction from MRI measurements, we will demonstrate that such workflows, along with hidden data preprocessing steps, can artificially inflate results by up to 40% and jeopardize clinical applicability. Next, we will provide hands-on demonstrations with open-access Python tools, where participants will learn to detect, quantify, and mitigate these sources of bias. The session equips the ISBI community with practical frameworks and guidelines for conducting transparent, reproducible, and trustworthy medical AI research.