What is MDAS?#
The Multimodal Data Analysis Session (MDAS): Cancer & Cardiac Research introduces practical workflows for working with different types of biomedical data. The Cancer and Cardiac libraries are built on PyKale.
Purpose and research focus#
Explore how multimodal machine learning can combine biomedical data types, and gain confidence adapting example workflows to research questions in cancer and cardiac applications.
Cancer: Combine clinical information and pathology patch features to explore time-to-event prediction with a synthetic cohort.
Cardiac: Explore ECG leads and chest X-ray/ECG fusion through synthetic-data demonstration workflows.
Your own research: Discuss which data modalities, outcomes and practical constraints matter in your project.
Who should attend?#
AI/ML researchers, data scientists, research software engineers, clinicians, clinical researchers and PhD students working with, or planning to work with, cancer or cardiac data.
Some Python experience and a laptop are recommended. Clinicians can focus on what data are used, what is predicted, how results are interpreted and where limitations arise. Technical participants can also inspect configurations and code.
Multimodal fusion and the analysis workflow#
The session introduces early, intermediate and late fusion at a high level before participants explore practical examples. Follow the workflow from data preparation and loading through encoding, fusion, prediction, evaluation and interpretation.
The Cancer track uses intermediate fusion in a Cox time-to-event model. The Cardiac track explores multimodal representation learning and lead-specific ECG workflows. Both start with synthetic data so participants can explore without a patient dataset.
What you will learn#
By the end of the session, participants should be better able to:
Identify relevant modalities and a research endpoint.
Follow a documented example end to end.
Distinguish a synthetic pipeline check from clinically meaningful evaluation.
Explain what a fusion strategy changes and what a reported metric means.
Identify the next steps and constraints for applying a workflow to their own data.
Session format and information#
Wednesday, 18 November 2026, 13:00–17:00.
Venue: Regent Court, University of Sheffield; The Diamond is a possible alternative. See the Session Agenda for venue details and confirmation arrangements.
Learn the Pipeline: introductions, environment setup and core content from data loading to model evaluation, followed by discussion.
Apply the Pipeline: make progress on a real research problem, participants’ own or an organiser-provided problem, with the AIRE team available to help.
See the Session Agenda for exact timings and the Participation Guide for how to prepare your question and record observations. Specific organiser-provided problems and final tutorial exercise selection remain to be confirmed.
Find your route#
Review the prerequisites and Quick Start, then choose a tutorial. Use the participation guide to record observations, questions and findings.
Important
The synthetic examples are demonstrations of software workflows, not evidence of clinical performance. Appropriate permissions and secure infrastructure are required for real clinical data.