What Substitution and Prediction Strategies Address the Challenge of an Unmeasurable C2-7 Cobb Angle?

Authors

Abstract

The C2-7 Cobb angle is an important parameter in evaluating cervical sagittal alignment, which is widely used for preoperative planning, identifying surgical indications, and postoperative assessment. However, this angle becomes unmeasurable in 28% to 49% of clinical radiographs because of poor visualization of the C7 inferior endplate, limiting treatment planning and radiographic follow-up in cervical alignment assessment. The C2-6 Cobb angle has been proposed as a substitute in previous research, but these studies were limited by small symptomatic cohorts from a single center and lacked both subgroup-specific and external validation. Furthermore, there is currently a lack of reference standards for the clinical use of the C2-6 Cobb angle, and no established machine-learning models are available to accurately predict the C2-7 Cobb angle. (1) Can the C2-6 Cobb angle serve as a reliable substitute for the C2-7 angle? (2) Can machine-learning models accurately predict the C2-7 Cobb angle? We conducted a retrospective, multicountry imaging study from January 2020 to January 2025, utilizing standing lateral cervical spine radiographs from a large hospital data set in China and public data sets from Vietnam and India. In China, 11,800 radiographs were initially screened. The inclusion criterion was cervical radiographs of sufficient clarity. The exclusion criterion was cervical radiographs with incomplete visualization of anatomic structures. Following these exclusions, 10,571 radiographs from China were included, comprising 10,000 standard standing lateral radiographs plus 284 implant and 287 flexion-extension radiographs. From the public data sets, 470 radiographs from Vietnam and 62 from India were reviewed, with no radiographs excluded. A total of 11,103 radiographs were available for final analysis. Key variables included demographics (age, sex), symptomatic status, implant status, and radiographic sagittal parameters derived from standing lateral views. Four orthopaedic specialists labeled keypoints on the original radiographs, including the corner points of C2 to C7 and the centroid of C2. An algorithm was employed for precise measurement of the C2-6 and C2-7 Cobb angles. The Pearson correlation coefficient was calculated to assess the strength of the correlation between the C2-6 and C2-7 Cobb angles, and a linear regression analysis was applied to derive a predictive equation for the C2-7 Cobb angle based on the C2-6 Cobb angle. Subsequently, the 10,000 standard Chinese standing lateral radiographs were randomly assigned to the training set (80%) and the testing set (20%). An independent validation set (n = 1103) was established to assess robustness, comprising 284 implant radiographs and 287 flexion-extension radiographs from China, together with 470 from Vietnam and 62 from India. Correlation analysis demonstrated a strong positive correlation between the C2-6 and C2-7 Cobb angles in the overall population (r = 0.92; p < 0.001). Machine-learning models incorporating the C2-6 Cobb angle and other sagittal parameters achieved high predictive accuracy for estimating the C2-7 Cobb angle, with Lasso regression performing best (R 2 = 0.93, mean absolute error [MAE] = 2.57). Additionally, strong performance was observed in the validation set (R 2 = 0.95, MAE = 3.21). In the subgroup analysis for the extension in males group, the linear model achieved the best validation results, with R 2 = 0.94 and MAE = 2.52. A strong correlation and high interpretable linear regression results between the C2-6 and C2-7 Cobb angles were observed across different countries, body positions, and implants, suggesting that the C2-6 Cobb angle can serve as a reliable substitute for the C2-7 Cobb angle in radiographic imaging. Further analysis revealed that the C2-6 Cobb angle is approximately 6° smaller than the C2-7 Cobb angle at the population level, which may serve as an important reference for standardized interpretation in clinical evaluation. Machine-learning models achieved high predictive accuracy for estimating the C2-7 Cobb angle, with the best performing model (Lasso regression) achieving an MAE of 2.57, offering an alternative clinical application option. To facilitate clinical use, we provide a freely available online tool ( http://c2-7cobbanglepredictionsystem.online ) that will be maintained for at least 15 years. Level III, diagnostic study.

Tags

Topics