Original Article
Comparing the Accuracy of Pose Estimation Methods and Radiographic Parameters in Adult Spinal Deformity Patients
2025 Volume 16 Issue 5 Pages 754-762
Details
2025 Volume 16 Issue 5 Pages 754-762
Introduction: Adult spinal deformity (ASD) is a condition characterized by complex three-dimensional spinal deformities that adversely affect the quality of life (QOL) in older adults. Current assessments for ASD rely primarily on standing X-ray imaging. However, this method presents challenges, such as radiation exposure and limited reproducibility due to posture variability.
This study aims to evaluate the utility of artificial intelligence (AI)-based pose estimation using RGB images, which refer to images where each pixel is defined by the intensity of red, green, and blue colors, as a non-invasive alternative to traditional radiographic assessments in patients with ASD.
Methods: A total of 23 patients with ASD underwent simultaneous standing full-spine X-ray and full-length lower limb X-ray imaging, along with RGB image capture. Using pose estimation AI, 17 anatomical key points were identified, and specific pose estimation parameters were defined. Correlations between these AI-derived parameters and those obtained from X-ray measurements were then analyzed.
Results: Significant correlations were observed between the pose estimation parameters and X-ray parameters. Specifically, in the coronal plane, shoulder balance, trunk tilt, and knee varus/valgus showed significant correlations, while in the sagittal plane, trunk tilt and knee flexion/extension parameters showed significant correlations. These results suggest that pose estimation AI, using RGB images, may serve as a viable, non-invasive alternative for assessing postural alignment in ASD patients.
Conclusions: The AI-driven pose estimation method used in this study demonstrates the potential to facilitate non-invasive and straightforward postural assessment for patients with ASD, even when clothed. These findings are promising for clinical applications, enabling more frequent assessments without the associated risks of radiation exposure. Further studies involving larger patient cohorts are recommended to improve the accuracy and reliability of this technology for broader clinical use.
