Related Experiment Video
Updated: Jun 21, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
385
Bidirectional Copy-Paste Mamba for Enhanced Semi-Supervised Segmentation of Transvaginal Uterine Ultrasound Images
Boyuan Peng1, Yiyang Liu1, Wenwen Wang2
1Graduate Department of Computer Science and Engineering, The University of Aizu, Aizu-Wakamatsu 965-8580, Japan.
Diagnostics (Basel, Switzerland)
|July 13, 2024
Summary
A new semi-supervised deep learning model, BCP-Mamba, improves parametrium segmentation in transvaginal ultrasound images. This method reduces the need for extensive manual annotation, aiding in uterine disease diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer-Aided Diagnosis
Background:
- Accurate parametrium segmentation in transvaginal ultrasound is crucial for diagnosing uterine diseases.
- Current fully supervised deep learning methods require costly, time-consuming pixel-level annotations.
Purpose of the Study:
- To develop a semi-supervised model for efficient parametrium segmentation.
- To reduce the reliance on extensive manual annotations in medical image analysis.
Main Methods:
- Introduction of a bidirectional copy-paste Mamba (BCP-Mamba) model.
- Integration of a U-shaped structure with a visual state space (VSS) module.
- Utilized a dataset of 1940 transvaginal ultrasound images.
Main Results:
- BCP-Mamba achieved a Dice coefficient of 86.55%, outperforming U-Net (80.72%) and BCP-Net (84.63%).
- The model demonstrated a lower Hausdorff_95 distance (14.56) compared to U-Net (23.10) and BCP-Net (21.34).
- The semi-supervised approach proved effective for transvaginal uterine ultrasound image segmentation.
Conclusions:
- The BCP-Mamba model offers a superior semi-supervised solution for parametrium segmentation.
- This approach can decrease the burden on expert sonographers.
- Facilitates more accurate prediction and diagnosis of uterine conditions.
