Sudden cardiac death (SCD) is the leading cause of mortality in athletes during sport participation. Resting 12-lead ECG is widely used for pre-participation screening. The International Criteria (IC) for ECG interpretation in athletes define which patterns of T-wave inversion (TWI) are considered abnormal and warrant further investigation to detect cardiomyopathies such as hypertrophic or arrhythmogenic cardiomyopathy. However, even when TWI is deemed abnormal, an underlying cardiomyopathy is identified in only about half of cases, making this a particularly challenging issue in sports cardiology. The lack of publicly labeled datasets makes AI-based ECG analysis in athletes rare; the process still relies on manual interpretation by specialists. A dataset labeling abnormal ECGs as benign or pathological, collected by a clinical partner between 2008 and 2025, is used here to introduce a clinically explainable ML pipeline built on features derived from the IC framework to distinguish pathological from physiological TWI. We studied 12-lead resting ECGs (500 Hz, 10 s) of 130 athletes presenting with abnormal TWI. Among them, 65 had clinically confirmed cardiac pathology, and 65 had benign athletic remodeling after comprehensive evaluation, including echocardiography, exercise testing, Holter monitoring, cardiac CMR, and follow-up. We extracted 475 handcrafted features from all 12 leads, aligned with the IC framework. Features were ranked using sensitivity-based selection and swept from 1 to 100 under paired leave-one-out cross-validation with several classifiers. The best result, using the top 80 features, achieved 86.2% accuracy to depict an underlying cardiomyopathy versus 50% for an experienced cardiologist. Because every feature maps to a clinical parameter cardiologists already use, predictions are traceable to interpretable ECG measurements — an important requirement for clinical trust. Interestingly, the T-wave axis alone yielded 73.3% accuracy, consistent with prior evidence linking T-axis deviation to cardiac risk, suggesting it may serve as a simple triage marker warranting prospective validation.