Researchers developed a modular framework that repurposes continuous diagnostic scores from electrocardiogram foundation models into interpretable classifiers for detecting left ventricular dysfunction. The approach converts foundation-model outputs into tree-based classification scores, aiming to preserve performance while improving clinical interpretability. The results describe enhanced benchmark performance for identifying low left ventricular ejection fraction, supporting a practical pathway for hospitals that already run ECG foundation models but need explainable risk stratification. The work highlights how model outputs can be engineered into clinician-facing decision support rather than relying on opaque end-to-end predictions.
Get the Daily Brief