Why the study?
Preparticipation cardiovascular screening in athletes requires ECG interpretation that demands expertise and can be resource-intensive. Artificial intelligence could support physicians by automating resting 12-lead ECG analysis during routine sports eligibility assessment.
Does a multimodal deep learning model for automated analysis of resting 12-lead ECGs and tabular clinical data accurately identify positive preparticipation cardiovascular screening outcomes compared to clinical evaluation in competitive athletes?
Observational (n=531)
No
Does a multimodal deep learning model for automated analysis of resting 12-lead ECGs and tabular clinical data accurately identify positive preparticipation cardiovascular screening outcomes compared to clinical evaluation in competitive athletes?
Automated deep learning-based analysis of 12-lead ECGs and clinical data during preparticipation screening in athletes is feasible and demonstrates moderate diagnostic discrimination (AUC 0.69) for identifying positive screening results.
Background Preparticipation cardiovascular screening (PPS) aims to prevent sudden cardiac death in athletes, but electrocardiogram (ECG) interpretation requires expertise and may be resource-intensive. Artificial intelligence (AI) could support physicians by enabling automated ECG analysis during routine sports eligibility assessment. Purpose To evaluate the feasibility and diagnostic performance of a deep learning (DL) model for automated analysis of resting 12-lead ECGs obtained during routine PPS in competitive athletes. Methods In this prospective monocentric observational study, competitive athletes aged 16–60 years undergoing routine PPS were consecutively enrolled. Screening included medical history, physical examination, resting ECG, and exercise testing. Athletes were classified as fit or not fit for competitive sports according to clinical evaluation. A multimodal DL architecture combining convolutional ECG feature extraction, temporal attention pooling, and tabular clinical data fusion was developed to classify PPS outcome. Model performance was assessed using stratified 10-fold cross-validation and reported as mean ± SD. Results A total of 531 athletes were included (72% male, mean age 32.2 years); 168 (32%) had a positive PPS result. In the independent test setting, the DL model achieved an accuracy of 0.62 ± 0.06, sensitivity 0.68 ± 0.18, specificity 0.56 ± 0.17, F1-score 0.71 ± 0.11, and area under the ROC curve (AUC) 0.69 ± 0.06, indicating moderate discrimination for identifying athletes with positive screening. Training performance was higher (accuracy 0.86 ± 0.06, AUC 0.96 ± 0.03), suggesting effective pattern learning but highlighting the need for broader validation. Conclusion Automated DL-based analysis of 12-lead ECGs during PPS is feasible and shows moderate diagnostic performance in independent testing. AI-assisted ECG interpretation may represent a useful adjunct to physician assessment for cardiovascular risk stratification in sports screening programs, warranting further validation in larger multicentre cohorts.
No takes yet. Share an insight, caveat, or question.
Marchetti et al. (2026) conducted an observational in Preparticipation cardiovascular screening in competitive athletes (n=531). Deep learning-based automated analysis of resting 12-lead ECG and clinical data vs. Clinical evaluation (medical history, physical examination, resting ECG, and exercise testing) was evaluated on Diagnostic performance for classifying preparticipation cardiovascular screening outcome (fit vs not fit for competitive sports) (AUC 0.69 ± 0.06).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: