Spinal Analysis Metrics and Dataset Evaluation for Scoliosis Detection

Cobb angle is often used to evaluate scoliosis, a condition of the spine marked by abnormal curvature. The development of automated systems is necessary because manual determination of the Cobb angle from X-ray images is prone to intraobserver and interobserver variability. This analysis presents a deep learning-based framework for accurate and reliable Cobb angle measurement. The proposed approach employs RetinaNet for vertebral detection, followed by coordinate extraction and curvature computation. Additional metrics, including Spine Regression Integral (SRI) and Spine Center Distance (SCD),), provide a comprehensive assessment of spinal alignment. The model is validated using the mean absolute error (MAE), Mean Average Precision ((mAP), and Intraclass Correlation Coefficient (ICC), achieving superior performance with an MAE of 1.845◦- 2.54◦ and a precision of 85%-90%. Compared to traditional manual methods and prior automated approaches, the proposed system demonstrates higher accuracy, robustness, and clinical applicability. This research contributes to scoliosis diagnosis by offering a scalable and interpretable deep learning solution, with potential for real-world clinical integration. Future work aims to improve model generalization and enhance performance in diverse imaging conditions.