Machine-Learning Algorithms to Automate Morphological and Functional Assessments in 2D Echocardiography

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Machine learning models can analyze cardiac tissue deformation for phenotypic recognition.
• Differentiating hypertrophic cardiomyopathy (HCM) from physiological hypertrophy in athletes (ATH) is clinically important.

Purpose of the Study:

• To evaluate a machine-learning framework using speckle-tracking echocardiography for automated discrimination between HCM and ATH.
• To assess the diagnostic value of machine learning in identifying pathological cardiac hypertrophy.

Main Methods:

• An ensemble machine-learning model was developed using support vector machines, random forests, and artificial neural networks.
• Expert-annotated echocardiographic data from 77 ATH and 62 HCM patients were utilized.
• K-fold cross-validation and majority voting were employed for model validation and prediction.

Main Results:

• Volume, mid-left ventricular segmental, and average longitudinal strain were key predictors identified by information gain.
• The machine-learning model demonstrated superior sensitivity and specificity compared to traditional echocardiographic parameters like E/e' and strain.
• In subgroup analysis of younger patients, the model maintained high sensitivity and improved specificity.

Conclusions:

• Machine learning algorithms can effectively discriminate between physiological and pathological cardiac hypertrophy.
• This study is a step towards developing real-time, automated echocardiographic interpretation systems.
• Such systems could aid clinicians, especially those with less experience, in diagnosing cardiac conditions.