Echocardiography-Based Deep Learning Model to Differentiate Constrictive Pericarditis and Restrictive Cardiomyopathy

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Constrictive pericarditis (CP) is a reversible cause of diastolic heart failure, but diagnosis is challenging.
• Artificial intelligence (AI) offers potential to improve CP identification.

Purpose of the Study:

• To develop and evaluate a deep learning model using transthoracic echocardiography to differentiate CP from cardiac amyloidosis (CA).

Main Methods:

• A ResNet50 deep learning model was trained on apical 4-chamber echocardiographic views from 381 patients (184 CP, 197 CA).
• Model performance was assessed using the area under the curve (AUC) on a held-out test set.
• GradCAM was employed for model interpretation.

Main Results:

• The AI model achieved an AUC of 0.97 in differentiating CP from CA.
• GradCAM analysis indicated model focus on the ventricular septal area.

Conclusions:

• An AI model utilizing standard echocardiography views can facilitate CP detection.
• This approach can enhance workflow efficiency and enable prompt patient referral for intervention.