M4S-Net: a motion-enhanced shape-aware semi-supervised network for echocardiography sequence segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Medicine

Background:

• Echocardiogram segmentation is crucial for diagnosing cardiovascular diseases but is challenged by low image quality and complex cardiac motion.
• Supervised learning methods are limited by the difficulty and cost of labeling echocardiography sequences.

Purpose of the Study:

• To develop a semi-supervised network (M4S-Net) for robust echocardiogram sequence segmentation.
• To address challenges posed by low image quality, complex motion, and limited labeled data.

Main Methods:

• Proposed M4S-Net incorporating multi-level shape priors for enhanced shape representation.
• Utilized a motion-enhanced optimization module with optical flows for geometric assistance and temporal consistency.
• Employed a hybrid loss function and parameter-sharing for semi-supervised sequence segmentation.

Main Results:

• M4S-Net demonstrated superior spatial and temporal segmentation performance compared to state-of-the-art methods on public and in-house datasets.
• Achieved an AUC of 0.944 on an apical rocking recognition task, surpassing specialized physicians.

Conclusions:

• M4S-Net effectively overcomes limitations in echocardiogram segmentation, offering improved accuracy and efficiency.
• The proposed method holds significant potential for advancing cardiovascular disease diagnosis and treatment.