Related Experiment Video
Updated: Jun 9, 2026
00:10
Precision Measurements and Parametric Models of Vertebral Endplates
Published on: September 17, 2019
6.5K
Synthetic 3D Spinal Vertebrae Reconstruction from Biplanar X-rays Utilizing Generative Adversarial Networks.
Babak Saravi1,2,3,4, Hamza Eren Guzel5, Alisia Zink2
1Department of Orthopedics and Trauma Surgery, Medical Center-University of Freiburg, Faculty of Medicine, University of Freiburg, 79106 Freiburg, Germany.
Journal of Personalized Medicine
|December 23, 2023
Summary
This study reconstructs 3D spinal vertebrae from X-rays using Generative Adversarial Networks (GANs), reducing radiation and cost. The AI model shows promise for enhanced spinal imaging diagnostics.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Healthcare
- Radiology
Background:
- Computed tomography (CT) provides detailed spinal anatomy but involves high radiation and cost.
- Conventional X-rays are lower cost and lower radiation but offer limited 3D anatomical detail.
- There is a need for advanced imaging techniques to improve spinal diagnostics while mitigating CT's drawbacks.
Purpose of the Study:
- To develop and evaluate a Generative Adversarial Network (GAN) framework for reconstructing 3D spinal vertebrae from biplanar X-ray images.
- To assess the feasibility of using synthetic X-ray data generated from CT segmentations for 3D reconstruction.
- To explore the potential of reducing radiation exposure and costs in spinal imaging.
Main Methods:
- Generated synthetic anterior and lateral X-ray images using the DRRGenerator module in 3D Slicer from CT-based spinal vertebrae segmentations.
- Applied a novel X2CT-GAN framework with feature fusion to reconstruct 3D spinal vertebrae from the synthetic biplanar X-rays.
- Trained the GAN generator using a combination of mean squared error (MSE) and adversarial loss.
Main Results:
- The GAN framework successfully reconstructed 3D spinal vertebrae CTs from synthetic biplanar X-rays.
- Quantitative evaluation using metrics such as PSNR (28.394 dB), PSNR-3D (27.432), SSIM (0.468), cosine similarity (0.484), MAE0 (0.034), and MAE (85.359) indicated effective reconstruction.
- Limitations were noted in capturing fine bone structures and precise vertebral morphology.
Conclusions:
- The proposed GAN-based approach effectively reconstructs 3D spinal vertebrae from biplanar X-rays, offering a potential alternative to traditional CT scans.
- This technique demonstrates the capability to enhance diagnostic insights from low-cost X-ray imaging while reducing radiation exposure and associated costs.
- Further refinement is needed to address limitations in fine structural detail for broader clinical application in spinal imaging.
