Deep-Learning-Based Disease Classification in Patients Undergoing Cine Cardiac MRI.

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Automated analysis of cardiovascular diseases from MRI can provide fast and reproducible clinical assessments.
• Deep learning (DL) offers potential for automated disease classification from medical imaging.

Purpose of the Study:

• To develop a DL algorithm for classifying cardiovascular diseases using MRI data.
• The algorithm aims to differentiate between normal subjects (NORM) and patients with dilated cardiomyopathy (DCM), hypertrophic cardiomyopathy (HCM), and ischemic heart disease (IHD).

Main Methods:

• A retrospective study involving 1337 subjects (568 NORM, 151 DCM, 177 HCM, 441 IHD).
• Bi-ventricular morphological, functional, and left ventricular strain features were automatically extracted from cine MRI.
• Variational autoencoder models were trained and evaluated using tenfold cross-validation.

Main Results:

• The DL algorithm achieved high Area Under the Curve (AUC) values: 0.952 (NORM), 0.881 (DCM), 0.908 (HCM), and 0.856 (IHD), with an overall accuracy of 0.778.
• Specificity for the normal class was 0.908, increasing to 0.961 with cotraining using unlabeled data.
• Strain features were identified as important for disease classification.

Conclusions:

• Automatically extracted cardiac function features from cine MRI show promise for disease classification, particularly for normal-abnormal differentiation.
• Strain features play a significant role in accurate disease labeling.
• Cotraining with unlabeled data can enhance the specificity of normal-abnormal classification.