Related Experiment Video
Updated: Nov 9, 2025
10:23
Author Spotlight: Three-Dimensional Cephalometric Landmark Annotation Demonstration on Human Cone Beam Computed Tomography Scans
Published on: September 8, 2023
3.3K
Deep learning approach for automatic landmark detection and alignment analysis in whole-spine lateral radiographs
Yu-Cheng Yeh1, Chi-Hung Weng2, Yu-Jui Huang1
1Department of Orthopaedic Surgery, Spine Division, Bone and Joint Research Center, Chang Gung Memorial Hospital and Chang Gung University College of Medicine, Taoyuan, Taiwan, ROC.
Scientific Reports
|April 8, 2021
Summary
Deep learning models accurately locate spinal landmarks and generate radiographic parameters from lateral spine X-rays. This artificial intelligence approach matches doctor reliability for most measurements, aiding spinal balance assessment.
Area of Science:
- Radiology
- Medical Imaging
- Artificial Intelligence
Background:
- Sagittal radiographic parameter measurement is crucial for human spinal balance assessment.
- Deep learning offers potential for automated landmark detection and alignment analysis in spinal imaging.
Purpose of the Study:
- To develop and evaluate deep learning models for automatic localization of 45 anatomic landmarks on whole-spine lateral radiographs.
- To generate 18 key radiographic parameters using these models.
- To compare the performance of the deep learning model against manual measurements and physician reliability.
Main Methods:
- Development of deep learning models trained on 2210 annotated whole-spine lateral radiographs.
- Automatic identification of 45 spinal anatomic landmarks.
- Generation of 18 radiographic parameters, including alignment and balance metrics.
- Statistical analysis to assess correlation with ground truth and comparison with human expert measurements.
Main Results:
- The deep learning models achieved high accuracy in localizing spinal anatomic landmarks.
- All generated radiographic parameters showed significant correlation with ground truth values (p < 0.001).
- The artificial intelligence system demonstrated reliability comparable to physicians for 15 out of 18 parameters.
- Localization accuracy and learning speed varied by spinal region, with cervical landmarks performing best.
Conclusions:
- The proposed deep learning system accurately localizes spinal landmarks and generates radiographic parameters for spinal balance assessment.
- The automated system shows favorable correlations with manual measurements and comparable reliability to physicians.
- This AI-driven approach has the potential to enhance the efficiency and consistency of spinal radiographic analysis.
