Automatic classification of heart failure based on Cine-CMR images

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Heart failure (HF) is a critical condition with high mortality.
• Accurate HF classification aids clinical diagnosis and treatment planning.
• Previous self-supervised learning (SSLHF) for HF classification on Cine-CMR lacked 4D data integration.

Purpose of the Study:

• To propose an automatic 4D heart failure classification algorithm.
• To enhance HF classification by integrating spatial and temporal information from Cine-CMR.
• To develop an advanced self-supervised learning framework for HF diagnosis.

Main Methods:

• Introduced the 4D-SSLHF framework, combining self-supervised image restoration and HF classification.
• Utilized three image transformation methods for enhanced spatial and temporal information exploration.
• Employed a Siamese Conv-LSTM network to integrate four-dimensional features simultaneously.

Main Results:

• Achieved an AUC of 0.8794 and ACC of 0.8402 on 184 patients via five-fold cross-validation.
• Demonstrated significant improvements over previous work, with AUC increasing by 2.89% and ACC by 1.94%.
• Validated the framework's effectiveness in accurately classifying different HF categories.

Conclusions:

• Proposed the novel 4D-SSLHF framework for HF classification using Cine-CMR.
• The 4D-SSLHF effectively mines 3D spatial and temporal information for accurate HF classification.
• The method shows potential in assisting physicians with personalized HF treatment strategies.