Semi-Supervised Echocardiography Video Segmentation via Adaptive Spatio-Temporal Tensor Semantic Awareness and Memory...

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Cardiology

Background:

• Accurate segmentation of cardiac structures in echocardiography is crucial for diagnosing heart disease.
• Challenges include speckle noise, low resolution, and sparse annotations, limiting current methods.
• Existing techniques often rely on computationally expensive optical flow or cross-frame attention.

Purpose of the Study:

• To develop an innovative echocardiography video segmentation framework.
• To address limitations of existing methods, particularly noise sensitivity and high computational costs.
• To improve segmentation accuracy and efficiency using inherent video correlations.

Main Methods:

• Exploits spatio-temporal correlation of echocardiography video feature tensors.
• Employs adaptive tensor singular value decomposition (t-SVD) in a learnable 3D transform domain.
• Introduces a memory flow method for inter-frame information propagation based on multi-scale affinities.

Main Results:

• Adaptive t-SVD reduces redundancy and enforces low-rank properties in feature tensors.
• The framework effectively captures temporal evolution using limited labeled frames.
• Memory flow method precisely resolves frame-to-frame variations, enhancing segmentation continuity and accuracy.

Conclusions:

• The proposed method demonstrates superior performance over state-of-the-art techniques.
• It effectively overcomes constraints of sparse annotations in echocardiography video segmentation.
• The framework offers a promising advancement for cardiac diagnosis through improved video analysis.