Automated Echocardiographic Detection of Heart Failure With Preserved Ejection Fraction Using Artificial Intelligence

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Detecting heart failure with preserved ejection fraction (HFpEF) is challenging due to discordant clinical and imaging features.
• Traditional diagnostic methods can be indeterminate, necessitating improved approaches.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) model for detecting HFpEF from transthoracic echocardiogram videos.
• To compare the AI model's performance against established clinical scoring systems.

Main Methods:

• A 3D convolutional neural network was trained on apical 4-chamber echocardiogram videos to classify HFpEF cases versus controls.
• The AI model's diagnostic outputs (HFpEF, no HFpEF, nondiagnostic) were evaluated on an independent dataset.
• Performance was compared to Heart Failure Association-Pretest Assessment, Echocardiographic and Natriuretic Peptide Score (HFA-PEFF) and Heavy, Hypertensive, Atrial Fibrillation, Pulmonary Hypertension, Elder, and Filling Pressure (H2FPEF) scores.

Main Results:

• The AI model demonstrated excellent discrimination in training (AUC 0.97) and validation (AUC 0.95) datasets.
• In independent testing, the AI model maintained high sensitivity (87.8%) and specificity (81.9%), outperforming clinical scores in reclassifying indeterminate cases.
• AI-identified HFpEF was associated with a nearly twofold increased mortality risk (HR 1.9).

Conclusions:

• An AI model utilizing a single echocardiogram video clip can effectively detect HFpEF.
• The AI approach offers superior performance compared to current clinical scores for HFpEF detection.
• This AI tool identifies patients at higher risk of mortality, aiding in risk stratification.