MAchine Learning to Boost the Early Diagnosis of Acute Cardiovascular Conditions

Current State of Research in the Field

Acute cardiovascular disease (ACVD) is the leading cause of death in Switzerland and Europe, responsible for 29% of deaths in Switzerland and 36% across Europe. The increasing prevalence of ACVD, including acute myocardial infarction (AMI), acute heart failure (AHF), pulmonary embolism (PE), and acute aortic syndromes (AAS), places a significant burden on healthcare systems. Diagnosing these conditions in emergency departments (EDs) is challenging due to overlapping symptoms and the need for rapid, accurate decision-making.

The introduction of cardiovascular biomarkers, including high-sensitivity cardiac troponin, B-type natriuretic peptide, and D-dimer has revolutionized early diagnosis. These biomarkers, alongside clinical assessments and electrocardiograms (ECGs), are now essential diagnostic tools. However, current diagnostic algorithms have still tremendous limitations.

Recent advances in machine learning (ML) and deep learning (DL) offer opportunities to improve diagnosis. ML-based ECG interpretation and deep transferable learning (DTL) techniques could enhance diagnostic accuracy by integrating complex ECG and biomarker data. AutoML approaches can further refine these models, reducing human error and improving clinical workflows.

The research team has conducted multiple large-scale studies leading to significant advancements in cardiovascular biomarker research and precision medicine. Their contributions include:

* Validation of the MI3 model, which uses ML to improve NSTEMI * Introduction of the BASEL ECG Score, a quantitative tool that enhances NSTEMI diagnosis. * Validation of CoDE-ACS, an ML-based clinical decision support-tool that predicts the probability of NSTEMI more effectively than standard cardiac troponin thresholds.

The team is now focussing on integrating ECG data with biomarkers using AI/ML to enhance accuracy and automate decision-making. Collaboration with international experts has enabled the successful application of neural networks to ECG interpretation. The next steps include:

* Refining ML-based ECG interpretation to incorporate non-additive effects. * Expanding ML models to include multiple cardiovascular conditions beyond AMI. * Integrating these AI-driven tools into clinical workflows and electronic health records.

This research aims to revolutionise cardiovascular diagnostics by leveraging AI and ML for more precise, faster, and clinically relevant decision-making.

Objectives:

1. Develop and implement a clinical decision support tool that visualizes key diagnostic data. 2. Train and validate ML models to diagnose acute cardiovascular diseases (ACVD). 3. Compare ML model performance with existing diagnostic algorithms. 4. Validate ML models in large international clinical trials. 5. Integrate ML models into the electronic patient record at the University Hospital Basel.