[Electrocardiogram Artificial Intelligence modeling]

The project aims to develop AI tools which are able to predict a variety of diagnoses utilizing raw ECG (electrocardiogram) inputs. These tools aid the physician with efficient, non-invasive decision support.

---

Objective

Developing accurate and reliable AI models that analyze raw ECG signals, which are able to predict a range of cardiac and non-cardiac outcomes. AI predictions will be provided in a dashboard for the cardiologists.

Methodology

Data collection: collecting a diverse set of raw ECG signals along with clinical data like diagnoses.

Model development: investigating and developing a foundation model which can serve for a diverse set of detection tasks.

A first project is focusing on an automated diagnosis for atrial fibrillation ECGs. Which aims to optimise the previously developed deep learning model for the detection of paroxysmal atrial fibrillation (see ADAM-project).

A second project is focusing on predicting the age from the ECG signal. The difference between deep-learning predicted age and the chronological age can give an indication on the health status of the patient.

In a third project we are developing a deep learning model for the detection of amyloidosis, a rare disease that causes amyloid to be deposited in the heart muscle. Early detection is crucial, as the disease can be treated but not cured.

Impact and future directions

Since ECG is a non-invasive measure, the AI tools will enable identification of cardiac irregularities or other diseases without intrusive procedures. These tools also offer the possibility of efficient screening, which can lead to timely interventions and cost savings in healthcare.

Currently, the focus lies on cardiac abnormalities, but in the future we also strive to predict non-cardiac outcomes.

General info and contact

Keywords (#): ECG, AI, non-invasive

RADar project research lead: Ir. Louise Vander Heyde, Ir. Koen Peynsaert

Principal investigator: MD Karl Dujardin

Status: Ongoing