Ultrasound Segmentation Using Semi-Supervised Learning: Application in Point-of-Care Sarcopenia Assessment.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Ultrasound imaging is vital for medical diagnostics, but automatic segmentation is difficult, especially with insufficient labeled data.
• Accurate segmentation is crucial for applications like sarcopenia assessment and emergency response.
• Existing methods struggle in scenarios with limited annotated ultrasound images.

Purpose of the Study:

• To develop a semi-supervised learning approach for enhancing ultrasound image segmentation accuracy.
• To leverage unlabeled ultrasound data to improve segmentation performance.
• To address the challenge of limited labeled data in medical image analysis.

Main Methods:

• Proposed a semi-supervised learning framework based on an encoder-decoder architecture.
• Incorporated contrastive learning techniques to utilize unlabeled data statistics.
• Collected and utilized a dataset of ultrasound images from patients and healthy volunteers.

Main Results:

• The proposed method demonstrated superior performance across various training data percentages (1% to 100%).
• Achieved comparable results to U-NET using only 10% of the labeled data.
• Significantly outperformed existing models like U-NET, CCT, and UniMatch in most training set splits.

Conclusions:

• The developed semi-supervised method is robust and efficient for ultrasound image segmentation.
• The approach is particularly effective in scenarios where labeled data is scarce.
• This work offers a promising solution for improving automated medical image analysis.