Related Experiment Video
Updated: Jun 27, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
A temporal enhanced semi-supervised training framework for needle segmentation in 3D ultrasound images
Mingwei Wen1, Pavel Shcherbakov2, Yang Xu1,3
1Department of Biomedical Engineering, College of Life Science and Technology, Huazhong University of Science and Technology, No 1037, Luoyu Road, Wuhan 430074, People's Republic of China.
Physics in Medicine and Biology
|April 29, 2024
Summary
This study introduces a novel semi-supervised framework for fast and accurate 3D ultrasound biopsy needle segmentation, improving accuracy with temporal information and reducing errors in kidney and prostate datasets.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Biomedical Engineering
Background:
- Automated biopsy needle segmentation in 3D ultrasound is crucial for navigation but challenging due to low resolution and interference.
- Current deep learning methods require extensive labeled data, struggle with real-time performance, and consume high memory.
Purpose of the Study:
- To develop a fast, accurate, and semi-supervised deep learning framework for 3D ultrasound biopsy needle segmentation.
- To leverage temporal information from sequential ultrasound images to improve segmentation accuracy without additional labeling costs.
Main Methods:
- Proposed a temporal information-based semi-supervised training framework incorporating a novel circle transformer module.
- Utilized consistency constraints for semi-supervision on unlabeled data.
- Trained the model using a combined loss function including cross-entropy, Dice similarity coefficient (DSC), and mean square error.
Main Results:
- Achieved superior performance compared to mainstream temporal segmentation and semi-supervised methods on kidney and prostate biopsy datasets.
- Demonstrated higher DSC (77.1% kidney, 78.5% prostate) and reduced needle tip position and length errors.
- Validated effectiveness using sequential images from beagle biopsy procedures.
Conclusions:
- The proposed method significantly enhances biopsy needle segmentation accuracy in 3D ultrasound.
- This approach offers a cost-effective way to improve biopsy navigation systems by utilizing temporal data.
- The framework addresses limitations of existing methods regarding data requirements and real-time performance.
