Helping AI Read the Heart More Reliably
PhD research by Jialu Tang investigates reliable ECG language models for cardiovascular care, improving trust, robustness, and clinical usability of AI driven healthcare systems.
, PhD candidate in the Department of Industrial Design at the Eindhoven University of Technology, defended her dissertation on 9 September 2026.
In her thesis, Towards Reliable Multimodal ECG Language Models for Cardiovascular Care, Tang examines how artificial intelligence can combine electrocardiogram data and language-based information to support cardiovascular care reliably and responsibly. Her work addresses a key challenge for healthcare systems worldwide. AI must deliver stable and verifiable diagnostic outputs in real clinical workflows.
Why It Matters
Cardiovascular diseases remain one of the leading causes of illness and death worldwide. Hospitals accumulate years of ECG recordings and clinical text for every patient, yet existing models treat signal interpretation as a static task and cannot capture disease progression across time. Clinicians increasingly rely on data-driven tools to analyze this stream, detect risks, and support decisions. This raises hard questions. Can these systems adapt to rare diseases and new patient populations from only a few examples? Do they still work when parts of a patient's record are missing? And how can their recommendations be interpreted and validated?
Tang's research addresses these questions by improving the reliability of multimodal ECG-language models, an emerging type of AI that combines heart signals with clinical text and health records to read, explain, and answer questions about the heart.
Looking Closer
Within the Design of Empowering Systems cluster, Tang developed models that go beyond classifying heart rhythms. Her models learn how to learn from past cases, so a handful of examples suffices for a new diagnostic task. They compare each ECG against the patient's earlier recordings to track change over time. They translate health record numbers into language, so one model reasons over signal and clinical context together. And they justify their reports and answers with catalogued ECG evidence or similar past cases retrieved from an archive.
Strong performance alone falls short in healthcare. Models must also be generalizable, robust, and transparent, because clinical decisions affect people's lives. Tang's designs deliver all three under difficult clinical conditions. Her models adapt to new tasks from a few labelled examples, keep working when records are incomplete, and ground their outputs in evidence that clinicians can verify. On real hospital records, across diagnostic, prognostic, and generative tasks, they improve robustness and generalisability over existing approaches.
Beyond the Lab
This work matters beyond hospitals and research laboratories. Entrepreneurs building digital health products face growing demands for explainability and accountability from regulators, healthcare organizations, and patients. New AI products must work, and they must also demonstrate that their outputs can be trusted.
Society is also learning how to live with AI that influences healthcare decisions. Patients want to know how recommendations arise, while clinicians need tools that perform consistently. Tang's work shows how AI can act as a diagnostic aid that supports clinicians' own decision-making.
Building Trust
Trust in healthcare AI must be earned. A system deserves that trust when it keeps performing as patients, records, and conditions change, and when its reasoning can be checked. Responsible innovation therefore requires more than technological advancement.
Tang's methods answer this need. They trace a single route, from models that adapt to new clinical situations, through models that reason over incomplete data, to models that justify their conclusions with verifiable evidence. Together they lay a stronger foundation for cardiovascular AI in clinical practice.
What Lies Ahead
Cardiovascular care is moving toward real-time risk evaluation and long-term patient monitoring, and multimodal AI will play a growing role in it. Tang's work points one way forward with models that cite similar past cases as evidence and stay up to date as new cases arrive, without retraining. These insights help researchers, developers, and healthcare organizations build systems that are both capable and dependable.
By improving the reliability of cardiovascular AI, this work helps make data-driven healthcare both effective and trustworthy.
Jialu Tang defended her thesis on 9 September 2026.
Title of the thesis:
Supervisors: Yuan Lu and Aaqib Saeed.