MoST-Net: Motion Memory Spatiotemporal Transformer Network for Myocardial Echocardiography Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Myocardial echocardiography segmentation is crucial for cardiac function assessment but is hindered by complex motion and image noise.
• Current memory-based methods struggle to capture temporal motion dynamics, and Transformer methods often require full image sequences, limiting real-time applications.

Purpose of the Study:

• To develop an advanced segmentation network, MoST-Net, for myocardial echocardiography that addresses limitations of existing methods.
• To improve the accuracy and robustness of myocardial segmentation, particularly in online and dynamic scenarios.

Main Methods:

• Proposed MoST-Net integrates motion modulation into long-term memory learning for continuous temporal dependency accumulation.
• Employed a memory prompt encoder and spatiotemporal decoder with a boundary uncertainty enhancement module.
• Utilized an adaptive ranking strategy to optimize memory length based on temporal consistency and segmentation quality.

Main Results:

• MoST-Net demonstrated superior performance compared to state-of-the-art methods on multiple datasets, even with significant noise.
• The network effectively captures short-term and long-range motion dynamics for improved segmentation accuracy.
• Analysis indicated that motion-aware memory learning reduces the need for excessively long memory banks.

Conclusions:

• MoST-Net offers a robust and effective solution for myocardial echocardiography segmentation, excelling in dynamic and noisy conditions.
• The proposed motion memory learning approach shows significant potential for real-time, online clinical applications in cardiac imaging.