Wavelet-Based Frequency Replacement and Edge Enhancement for Semi-Supervised Fetal Ultrasound Image Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Ultrasound image segmentation is challenging due to low contrast and blurred boundaries.
• Fully supervised methods need extensive annotations, which are costly and time-consuming.
• Developing semi-supervised methods is crucial for limited annotation scenarios.

Purpose of the Study:

• To develop an effective semi-supervised segmentation framework for ultrasound images with limited annotations.
• To improve segmentation accuracy in ultrasound imaging by leveraging data augmentation and edge enhancement techniques.

Main Methods:

• Proposed a novel semi-supervised segmentation framework for ultrasound images.
• Utilized frequency component augmentation via discrete wavelet transform (DWT).
• Implemented an edge mask enhancement module to emphasize anatomical boundaries.

Main Results:

• Achieved average Dice similarity coefficients (DSC) of 0.81 and 0.91 on public fetal ultrasound datasets with only 10 annotated images.
• Demonstrated a 2-3% DSC improvement over existing semi-supervised methods like FixMatch.
• Ablation studies confirmed the effectiveness of high-frequency augmentation and edge enhancement.

Conclusions:

• The proposed framework effectively improves ultrasound image segmentation accuracy with limited annotations.
• Combines frequency-domain augmentation and edge-aware enhancement for robust performance.
• Offers a promising direction for medical image analysis in data-scarce environments.