TWIN-X: Multimodal Digital Twins with Generative AI for Explainable Precision Medicine

Background

Diagnosing and monitoring patients depends on many different kinds of data: medical images, laboratory results, health risk factors, and clinical reports from consultations. Bringing this information together and drawing meaningful conclusions from it is time-consuming, and takes time away from direct patient contact.

At the same time, health data is often fragmented across systems. This can delay follow-up examinations and interventions, with potentially serious consequences for patients at risk.

Digital twins offer a promising new approach. By combining imaging, lab tests, patient records, and other sources, AI can build a virtual representation of an individual patient and simulate how their disease may develop or respond to treatment. This opens the door to more accurate diagnoses, more personalised treatment decisions, and a better understanding of complex conditions such as cancer and cardiovascular disease.

Aim

TWIN-X aims to develop a trustworthy and explainable multimodal AI model that integrates diverse health data into a comprehensive digital twin of the patient. Our goals are to:

Integrate heterogeneous health data (imaging, laboratory results, clinical reports, and risk factors) into a single patient-level digital twin.
Use generative AI to model and simulate disease progression and treatment response.
Ensure the resulting models are explainable and trustworthy, so they can support clinical decision-making in cardiovascular disease and oncology.

At DIAG, Radboudumc acts as a clinical validation partner across multiple work packages, building on its clinical expertise in cardiovascular imaging and its infrastructure for real-world data collection and validation. Our contributions include:

Developing radiology report templates and validating text-structuring pipelines on cardiovascular cohorts.
Validating radiology segmentation and embedding models on coronary CT, aortic CT, and cardiac MRI data, including expert reader feedback and testing robustness across scanner vendors.
Providing clinical acceptance criteria and validation datasets for integrating social determinants of health.
Acting as an early adopter of the integrated digital twin framework, running site acceptance tests on cardiology pathways.
Contributing cardiovascular imaging cohorts and adjudicating clinical endpoints for the cardiology demonstrator study on cardiomyopathy and coronary artery disease.
Participating in human factors studies and contributing clinical faculty to the consortium training programme.

The project started in June 2026 and runs for 48 months. More information is available on the TWIN-X website.