Machine Learning Analysis of Left Ventricular Function to Characterize Heart Failure With Preserved Ejection Fraction

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Current diagnostic methods for heart failure with preserved ejection fraction (HFpEF) are suboptimal.
• Objective assessment of left ventricular (LV) function during rest and exercise is crucial for understanding HFpEF.
• Machine learning (ML) offers potential for advanced analysis of complex physiological data.

Purpose of the Study:

• To test if comprehensive ML analysis of LV function during exercise can objectively differentiate HFpEF from healthy subjects.
• To explore the continuum of cardiac function from health to HFpEF using ML.
• To validate ML-derived diagnostic zones against current clinical criteria.

Main Methods:

• Unsupervised ML algorithm applied to left ventricular long-axis myocardial velocity patterns from stress echocardiography.
• Analysis of 156 subjects (HFpEF, healthy, hypertensive, breathless) in the MEDIA study.
• Clinical validation and independent evaluation of ML algorithm performance.

Main Results:

• ML identified a continuum from health to disease, including a transition zone for uncertain diagnoses.
• ML-diagnostic zones showed significant differences in clinical parameters (age, BMI, 6MWD, BNP, LVMI).
• ML accurately correlated with clinical diagnosis (κ=72.6%) and identified previously undetected abnormalities.

Conclusions:

• Interpretable ML analysis of LV function during exercise shows promise for improving HFpEF diagnosis.
• ML can reveal subtle differences in cardiac function, aiding in the understanding of HFpEF.
• This approach may refine diagnostic criteria and identify mild forms of HFpEF in at-risk populations.