High-Throughput Precision Phenotyping of Left Ventricular Hypertrophy With Cardiovascular Deep Learning

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Early detection of increased left ventricular (LV) wall thickness is crucial but challenged by under-recognition, measurement errors, and difficulty distinguishing causes.
• Conditions like hypertrophic cardiomyopathy and cardiac amyloidosis require accurate differentiation for effective patient management.

Purpose of the Study:

• To evaluate the accuracy of a deep learning (DL) workflow in quantifying ventricular hypertrophy.
• To assess the DL model's ability to predict the causes of increased LV wall thickness, specifically hypertrophic cardiomyopathy and cardiac amyloidosis.

Main Methods:

• A cohort study utilizing physician-curated data from specialized cardiac clinics (Stanford and Cedars-Sinai) between 2008 and 2020.
• A DL algorithm trained and validated on retrospective echocardiogram videos from multiple healthcare systems, including independent external datasets.
• Analysis focused on the DL model's accuracy in measuring LV dimensions and classifying specific causes of hypertrophy.

Main Results:

• The DL algorithm demonstrated high accuracy in measuring intraventricular wall thickness (MAE 1.2 mm) and LV diameter (MAE 2.4 mm).
• It accurately classified cardiac amyloidosis (AUC 0.83) and hypertrophic cardiomyopathy (AUC 0.98) in the primary cohort.
• External validation confirmed accurate quantification of ventricular parameters (R2 0.96 domestic, R2 0.90 international) and detection of hypertrophic cardiomyopathy (AUC 0.89) and cardiac amyloidosis (AUC 0.79).

Conclusions:

• The DL model accurately identifies subtle LV wall geometric changes and differentiates causes of hypertrophy.
• The fully automated DL workflow provides reproducible, precise measurements, surpassing human expert variability.
• This technology may form the basis for a precision diagnostic approach to cardiac hypertrophy.