Deep-Learning Models for the Echocardiographic Assessment of Diastolic Dysfunction.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Clinical algorithms for phenotyping diastolic dysfunction in heart failure with preserved ejection fraction (HFpEF) lack precision.
• Accurate patient stratification is crucial for effective HFpEF management.

Purpose of the Study:

• To develop and validate a deep neural network (DeepNN) model for identifying distinct HFpEF patient subgroups.
• To assess the model's ability to predict elevated left ventricular filling pressure and clinical outcomes.

Main Methods:

• A DeepNN model was developed using multidimensional echocardiographic data from a derivation cohort (n=1,242).
• Model performance was validated in external cohorts for predicting elevated left ventricular filling pressure and prognostic value.
• Clinical significance was assessed in HFpEF trials (TOPCAT, NEAT-HFpEF, RELAX-HF) by correlating phenogroups with outcomes, biomarkers, and exercise parameters.

Main Results:

• The DeepNN model outperformed current guidelines in predicting elevated left ventricular filling pressure (AUC 0.88 vs. 0.67).
• High-risk phenogroups demonstrated significantly higher rates of heart failure hospitalization/death and better response to spironolactone.
• High-risk patients exhibited increased myocardial injury, neurohormonal activation, and reduced exercise capacity.

Conclusions:

• The DeepNN classifier accurately characterizes diastolic dysfunction severity and identifies HFpEF subgroups with elevated filling pressures.
• This tool aids in identifying patients at high risk for adverse events and those likely to benefit from spironolactone therapy.
• The model offers a publicly available, precise method for HFpEF phenotyping and risk stratification.