AI-driven electromechanical model for non-invasive CAD risk assessment across symptom profiles

Introduction Accurate pre-test risk stratification for obstructive coronary artery disease (CAD) remains a challenge in primary care, particularly among patients with no symptoms. Traditional guideline-based models such as the ESC 2019 pre-test probability (PTP) tool are limited to symptomatic patients and frequently categorise a large proportion as intermediate risk, complicating decision-making and increasing unnecessary referrals. Purpose To evaluate the diagnostic performance of a novel Electro-Mechanical Risk (EMR) Score—a deep learning model that integrates seismocardiographic (SCG) features and clinical variables—for non-invasive CAD risk estimation. Performance was compared with the ESC 2019 PTP and risk factor–clinical likelihood (RF-CL) models in symptomatic patients and assessed independently in asymptomatic individuals. Methods A deep learning model based on a one-dimensional convolutional neural network (1D CNN) was trained on SCG-derived time-frequency features and clinical inputs, including age, sex, chest pain type, and cardiovascular risk factors. Ground truth was defined by coronary CT or invasive angiography. Diagnostic performance was evaluated using AUC, sensitivity, specificity, and category-based reclassification. Risk thresholds were set at 15% for the EMR Score and 5% for ESC and RF-CL models. Results In a multicentre cohort of 2,110 patients (32% CAD prevalence), the EMR Score achieved an AUC of 0.90 overall. In symptomatic patients, EMR outperformed the ESC 2019 model (AUC = 0.88 vs 0.83) and RF-CL (AUC = 0.88 vs 0.85), with greater separation between CAD and non-CAD cases across low-, intermediate-, and high-risk groups. EMR reduced the proportion of intermediate-risk patients by 33%, reclassifying 17% as low-risk and 16% as high-risk. Among those reclassified to low-risk, the negative predictive value was 91%. In asymptomatic patients, where ESC 2019 does not apply, EMR maintained high discrimination (AUC = 0.89) and strong rule-out performance (NPV = 99%). Conclusion(s) The EMR Score significantly improves CAD risk stratification over guideline-based models in symptomatic patients, reducing diagnostic ambiguity and enabling more confident clinical decisions. Its applicability to asymptomatic patients further supports its use as a scalable, AI-powered tool for early detection of CAD across diverse outpatient settings.