Abstract
BACKGROUND: Heart failure with preserved ejection fraction (HFpEF) remains challenging to diagnose due to the complexity of diastolic function assessment during stress echocardiography, where multiple hemodynamic parameters must be evaluated under time pressure. Explainable artificial intelligence, specifically rule-based Clinical Decision Support Systems (CDSS), offers promising improvements in reproducibility and interpretability. METHODS: A rule-based CDSS was developed and clinically validated to automate left ventricular diastolic function assessment during semi-supine bicycle stress echocardiography. A prospective cohort of 134 patients (mean age 61.3 ± 8.7 years) with exertional dyspnea and preserved left ventricular ejection fraction (LVEF >50%) was enrolled, excluding individuals with significant valvular pathologies, arrhythmias, or unstable ischemia. Echocardiographic and Doppler data were collected using Toshiba Aplio500 and Esaote MyLabSIGMA systems. The algorithm incorporated manual input of measurements, computed derived indices (e.g., diastolic reserve index, myocardial stiffness, vascular resistance), and applied rule-based logic in accordance with ASE/EACVI (2016/2022) guidelines and the ESC HFpEF consensus. RESULTS: The CDSS generated diagnostic conclusions within 3 min per case, matching expert assessments in 93% of cases and correctly identifying stress-induced diastolic dysfunction in 85%. It demonstrated high diagnostic agreement (ICC > 0.94) and discrimination (AUC = 0.92). Rule-based outputs, such as "Impaired diastolic reserve" or "Right ventricular dysfunction under load," were based on combinations of parameters (e.g., E/e' > 15, Δe' ≤ 0, TAPSE < 17 mm, PCWR > 12 mmHg). CONCLUSION: The explainable, guideline-compliant CDSS enables real-time, transparent analysis of diastolic function, supporting improved diagnostic consistency and augmented physician decision-making in cardiovascular care.