P-006 Radiology & imaging Poster

Beyond diagnostic accuracy: interpretable ECG-based machine learning for differentiating acute myocarditis and acute coronary syndrome

Audience Choice Award

To rate this poster, scan the QR code on the poster itself — voting only works on site. Every attendee has 3 stars and can give at most one per poster. A star is final and cannot be taken back.

Authors

W. Lauth presenting

Team Biostatistics and Big Medical Data, IDA Lab Salzburg, Paracelsus Medical University Salzburg, Salzburg, Austria

#0
M. Mirna

Division of Cardiology, Department of Internal Medicine II, Paracelsus Medical University of Salzburg, Salzburg, Austria

#1
G. Zimmermann

Department of Artificial Intelligence and Human Interfaces, Paris Lodron University of Salzburg, Salzburg, Austria

#2

Abstract

Acute myocarditis and acute coronary syndrome (ACS) often present with similar clinical features, including new-onset chest pain and elevated troponin levels, making differential diagnosis challenging. Although coronary angiography is commonly used to exclude ACS in patients with suspected myocarditis, it is invasive and resource-intensive. Electrocardiography (ECG), in contrast, is non-invasive, inexpensive, and widely available, but its ability to reliably distinguish between these conditions remains limited. This study aimed to develop and evaluate ECG-based machine learning models for differentiating acute myocarditis from ACS.

ECGs from 178 patients (86 with myocarditis and 92 with ACS) collected during routine clinical care at a single center were used to train predictive models. Two approaches were investigated: a Random Forest classifier representing traditional machine learning and a Convolutional Neural Network (CNN) representing deep learning. Models were trained to classify myocarditis and ACS directly from ECG data and evaluated using patient-wise five-fold cross-validation. Performance was assessed using area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity.

To assess model interpretability, feature importance analysis was applied to the Random Forest model, while Gradient-weighted Class Activation Mapping (Grad-CAM) was used to identify ECG regions influencing CNN predictions. Preliminary findings showed strong classification performance for the Random Forest model; however, interpretability analyses suggested reliance on potentially non-physiological features. CNN performance and interpretability analyses are ongoing.

These results demonstrate the potential of ECG-based machine learning approaches to support differentiation between acute myocarditis and ACS while emphasizing the importance of interpretability for ensuring clinically meaningful model behavior. Such approaches may ultimately provide a readily available, less invasive diagnostic aid in clinical practice.