Related Experiment Video
Updated: Aug 6, 2026
06:45
Automated Joint Space Detection Improves Bone Segmentation Accuracy
Published on: November 28, 2025
Intelligent recognition and segmentation of anatomical structures in spinal endoscopy: a deep learning approach with
Jingtian Yuan1, Junwei Zhang2, Tairui Zhang3
1Baikal School of BRICS, Irkutsk National Research Technical University, Irkutsk, Russian Federation.
Summary
This study introduces an AI system for segmenting spinal endoscopic images, accurately identifying key structures like the ligamentum flavum and nerve roots. While effective for soft tissues, further improvements are needed for bone and ligament segmentation for clinical navigation.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Spinal Surgery
Background:
- Accurate intraoperative identification of anatomical structures is crucial for safe and effective spinal endoscopic surgery.
- Challenges include limited visual fields, image blurring, and complex anatomy, hindering existing intelligent systems.
- Robust multi-structure segmentation is needed for computer-aided surgical navigation.
Purpose of the Study:
- Develop and validate a deep learning system for simultaneous segmentation of multiple anatomical structures in spinal endoscopic images.
- Provide a reliable foundation for computer-aided surgical navigation in spinal procedures.
- Enhance the accuracy and robustness of automated segmentation in challenging endoscopic environments.
Main Methods:
- Constructed a large-scale dataset of 1000 expert-annotated spinal endoscopic images.
- Developed an enhanced U-Net model integrating Convolutional Block Attention Module (CBAM) and Atrous Spatial Pyramid Pooling (ASPP).
- Evaluated model performance using five-fold cross-validation and an independent test set, comparing against benchmark algorithms.
Main Results:
- Achieved high segmentation performance for ligamentum flavum (Dice: 0.882), nerve roots (mIoU: 0.800), and intervertebral disc (mIoU: 0.723).
- The integrated CBAM and ASPP modules significantly contributed to performance.
- Outperformed baseline and state-of-the-art models, showing a 17.0% relative improvement in mIoU for ligamentum flavum compared to DeepLabV3+.
Conclusions:
- The developed system demonstrates high accuracy for segmenting critical soft tissues like the ligamentum flavum and nerve roots in spinal endoscopic images.
- Integration of attention mechanisms and multi-scale feature extraction is an effective strategy for improving segmentation.
- Further refinement is needed for accurate segmentation of the posterior longitudinal ligament and bone before clinical application in real-time navigation.