Validation of an aAI-supported vectorcardiography for the detection of coronary artery disease: a comparative study with coronary calcium scoring and CT Angiography (AI-CCC Study)

Background/Introduction Coronary artery disease (CAD) remains a leading cause of global mortality. Although computed tomography angiography (CTA) and calcium scoring (CS) are well-established diagnostic methods, they require specialised equipment and involve radiation and contrast exposure. Vectorcardiography (VCG) is a radiation-free alternative, and recent advances in artificial intelligence (AI) have led to the development of an AI-based 5-lead 3D VCG (5L3DVCG-AI) system that is capable of extracting complex electrophysiological indices to non-invasively characterise myocardial ischaemia. Purpose This study evaluates the diagnostic and ordinal staging utility of the 5L3DVCG-AI's CSG-index (a composite measure of cardiac spatial geometry) and p-factor (reflecting repolarisation dispersion) for detecting CAD, using CTA as the reference standard, and compares their performance alongside the Agatston Calcium Score. Methods In this observational study, 118 consecutive outpatients (mean age 52.4 years; 47% female) evaluated for suspected CAD underwent 5L3DVCG-AI, non-contrast CT for CS, and CTA. The coronary stenosis was categorised by CTA as follows: Grade 0 (no stenosis, n=71), Grade 1 (<50%, n=26), and Grade 2 (≥50%, n=21). The CSG-index and p-factor were analysed across groups. Diagnostic performance was assessed using receiver operating characteristic (ROC) analysis. Given the non-normal distribution of the predictors, non-parametric tests (Kruskal-Wallis, Wilcoxon) and Tukey HSD post-hoc analyses were utilised. To address data separation, Firth penalised logistic regression and 5-fold cross-validated Ridge regression were applied to assess independent predictive value. Results The CSG-index demonstrated a significant stepwise increase across stenosis severity groups (p<0.001), with mean values of −0.84 (Grade 0), −0.18 (Grade 1), and +0.07 (Grade 2). Post-hoc analysis showed that the CSG-index significantly differentiated between all three groups, including discrimination between the Grade 1 and Grade 2 (p=0.029). While CS was the strongest standalone predictor (AUC 0.971), the CSG-index (AUC 0.817) and p-factor (AUC 0.811) demonstrated strong independent diagnostic discrimination for coronary stenosis. Firth penalised regression confirmed that both the CSG-index (p=0.002) and p-factor (p<0.001) provided statistically significant additive predictive value beyond the CS. A model integrating CS, CSG-index, and p-factor achieved an AUC of 0.977 (sensitivity 95.7%, specificity 87.1%) Conclusion(s) The CSG-Index and p-factor demonstrate significant associations with coronary stenosis and offer high diagnostic accuracy. While CS remains the dominant predictor of plaque burden, 5L3DVCG-AI demonstrates strong diagnostic discrimination as a non-invasive, radiation-free modality. Its utilisation may facilitate the optimisation of patient selection for advanced imaging or intervention.ROC Derived Diagnostics TableCSG-Index Distrubition