A Self-supervised Learning Based Framework for Automatic Heart Failure Classification on Cine Cardiac Magnetic...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Heart failure (HF) is a critical condition with high mortality rates.
• Accurate classification of HF based on left ventricular ejection fraction (EF) is vital for effective clinical treatment.
• Cine cardiac magnetic resonance imaging (Cine-CMR) offers more precise EF estimation than echocardiography, yet its application in HF classification remains underexplored.

Purpose of the Study:

• To propose a self-supervised learning framework (SSLHF) for automated HF classification using Cine-CMR.
• To enable the classification network to effectively learn spatial and temporal information from Cine-CMR data.
• To classify HF patients into preserved EF and reduced EF categories.

Main Methods:

• A two-stage self-supervised learning framework (SSLHF) was developed.
• Stage 1: Self-supervised image restoration using a U-Net like network to extract HF-related spatial and temporal information.
• Stage 2: HF classification using a network initialized with weights from the Stage 1 encoder.

Main Results:

• The SSLHF framework achieved an Area Under the Curve (AUC) of 0.8505.
• The framework attained an Accuracy (ACC) of 0.8208 in 5-fold cross-validation.
• The self-supervised pre-training significantly enhanced the model's ability to classify HF patients.

Conclusions:

• The proposed SSLHF framework demonstrates a promising approach for accurate HF classification using Cine-CMR.
• Self-supervised learning effectively leverages spatial and temporal information in medical images for improved diagnostic accuracy.
• SSLHF offers a valuable tool for clinical decision-making in heart failure management.