Cardiovascular Diagnosis Classifier with Explainable AI (ICVD-XAI)

Cardiovascular diseases are among the leading causes of death worldwide and affect millions of people every year. Early detection of risk factors and transparent medical diagnostics are crucial for preventing severe disease progression and premature death. The electrocardiogram (ECG) plays a central role in this context, as it is a noninvasive and widely used diagnostic method that provides important information about the heart’s electrical activity.

Modern data-driven approaches—particularly machine learning and deep learning methods—open up new possibilities for automatically analyzing ECG data and reliably identifying clinically relevant patterns. At the same time, the explainability of such systems is becoming increasingly important to ensure trust and acceptance among medical professionals.

This is precisely where our project comes in: With the Interactive Cardiovascular Diagnoses Classifier with Explainable AI (ICVD-XAI), we are developing an innovative system that not only automatically classifies ECG data but also visualizes these decisions in a transparent and understandable way. The project combines state-of-the-art AI technologies with the practical requirements of clinical diagnostics—making a lasting contribution to safer and more transparent cardiovascular care.

The analysis of ECG signals faces several challenges: The data is prone to noise, varies depending on the recording method, and includes different signal formats ranging from 1-channel to 12-channel ECGs. Despite significant advances in modern AI models, the low explainability of many methods in particular remains an obstacle to their clinical application. While “explainable AI” methods such as SHAP, LIME, GradCAM, and LRP already exist, there is a lack of integrated systems that can analyze and visually present multiple ECG evaluations per patient in an explainable manner.

Added to this is a practical problem: In many cases, ECG findings are still available only in paper form. The digitization of these printouts has so far been unreliable, as recent studies and competitions have shown. This makes it considerably more difficult to compare ECG trends over several days or treatment phases.

The Interactive Cardiovascular Diagnoses Classifier with Explainable AI (ICVD-XAI) project aims to develop an innovative, interactive dashboard that presents ECG analyses in a transparent, understandable, and clinically useful way using explainable AI methods. To this end, various XAI approaches—such as SHAP, LIME, GradCAM, and LRP—will be integrated to make diagnostic decisions traceable. Combining multiple ECG recordings per patient into a single analysis dashboard represents another innovative approach. The project is being carried out in collaboration with the University of Applied Sciences of Graubünden, the Davos High-Altitude Clinic, and Cardio-Care. Together, the project aims to pool scientific, technical, and medical expertise to develop trustworthy AI systems for clinical use. The project makes an important contribution to the development of transparent, clinically applicable AI diagnostic systems while also strengthening regional research and healthcare expertise in the canton of Graubünden. It lays the foundation for future research in the fields of ECG digitization and AI-assisted diagnostics.

The project have also been supported by the following individuals:

• Prof. Dr. Andreas Ziegler – Scientific Director and CEO, Cardio-Care
• PD Dr. med. David Niederseer – Chief Physician of Cardiology, Hochgebirgsklinik Davos

Parties involved

The project is being carried out by the Institute for Data Analysis, Artificial Intelligence, Visualization, and Simulation (DAViS).