Related Experiment Video
Updated: Mar 18, 2026
09:29
Author Spotlight: Improving Percutaneous Vertebroplasty Surgical Accuracy and Efficiency Through Advanced Puncture Techniques
Published on: August 9, 2024
683
Boundary sensitive-net-based lumbar vertebra segmentation and spondylolisthesis measurement
Dongsheng Ji1, Furao Qian2, Yan Zong2
1School of Computer and Communication, Lanzhou University of Technology, 36 Pengjiaping Road, Lanzhou, 730050, Gansu, China. jids@lut.edu.cn.
Scientific Reports
|March 17, 2026
Summary
A new deep learning model, Boundary-Sensitive Network (BS-Net), accurately segments and quantifies lumbar vertebrae. This offers an efficient solution for diagnosing lumbar spondylolisthesis, improving upon traditional methods.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Spine Surgery
Background:
- Lumbar spine disorders are a major health issue.
- Accurate diagnosis requires precise vertebral segmentation and quantification.
- Existing methods like Cobb angle measurement and automated CT analysis have limitations.
Purpose of the Study:
- To develop an advanced deep learning framework for automated lumbar vertebral segmentation and quantification.
- To enhance the accuracy and efficiency of diagnosing lumbar spondylolisthesis.
Main Methods:
- Proposed a Boundary-Sensitive Network (BS-Net) incorporating Multi-Task Edge Processing (MEP) and Contextual Bilateral Fusion (CBF) modules.
- Integrated edge loss functions with morphological post-processing for joint segmentation and quantification.
- Evaluated on lumbar CT images and the SPIDER MRI dataset.
Main Results:
- BS-Net achieved a Mean Intersection over Union (MIoU) of 96.56% and a Dice coefficient of 98.5%.
- Spondylolisthesis quantification demonstrated strong agreement with manual assessments (Intraclass Correlation Coefficient > 0.9).
- Outperformed baseline models in segmentation and quantification tasks.
Conclusions:
- BS-Net offers an efficient and accurate automated solution for lumbar spondylolisthesis diagnosis.
- The model shows significant clinical value in improving diagnostic accuracy.
- Deep learning advancements can overcome limitations of traditional spine imaging analysis.
