Temporal Features-Fused Vision Retentive Network for Echocardiography Image Segmentation

Area of Science:

• Medical Imaging
• Cardiology
• Artificial Intelligence

Background:

• Echocardiography is crucial for assessing cardiac function by measuring left ventricular volumes at end-diastole (ED) and end-systole (ES) frames.
• Current methods often overlook correlations between consecutive echocardiography frames, potentially limiting accuracy.

Purpose of the Study:

• To develop an advanced model for echocardiography image analysis that leverages inter-frame correlations and spatial priors.
• To improve the accuracy of semantic segmentation for enhanced cardiac function assessment.

Main Methods:

• An encoder-decoder architecture incorporating a Temporal Feature Fusion Module (TFFA) using self-attention for inter-frame correlation.
• A Vision Retentive Network (Vision RetNet) encoder that integrates spatial priors via a Manhattan distance-based spatial decay matrix.
• Model evaluation on the EchoNet-Dynamic and CAMUS datasets.

Main Results:

• The proposed model demonstrated competitive performance on benchmark datasets.
• Incorporating spatial prior information and inter-frame correlations significantly enhanced semantic segmentation accuracy.
• The effectiveness of inter-frame correlations was amplified when spatial priors were also utilized.

Conclusions:

• Spatial priors and inter-frame correlations are vital for improving semantic segmentation in echocardiography.
• The developed model offers a promising approach for more accurate cardiac function assessment using echocardiography.
• Future research can build upon these findings to further refine cardiac imaging analysis.