Reliability analysis of automated Cobb angle measurement using artificial intelligence models in scoliosis patients

The measurement of the Cobb angle is essential for diagnosing scoliosis and monitoring its pro-gression, especially in adolescent idiopathic scoliosis. This study evaluates the reliability and efficiency of DEEP:SPINE-AS-01® developed by DEEPNOID, a Korean Ministry of Food and Drug Safety certified AI software, in automated Cobb angle measurement compared to manual methods. Fifty-two radiographs were initially collected, and 48 met inclusion criteria for analysis. Two observers, one an experienced spine orthopedic specialist and the other a non-specialist, performed manual measurements, while the AI module provided automated assessments. Intra- and inter-observer reliability were analyzed using intraclass correlation coefficients. The AI demonstrated excellent agreement with manual measurements, with ICCs above 0.98 for inter-observer comparisons. The non-specialist observer required more time for manual measurements (mean 53 minutes) than the specialist (mean 36 minutes), highlighting the manual method’s dependence on experi-ence. DEEP:SPINE-AS-01® reduced measurement variability and time, offering results in de-grees with high reproducibility. Despite minor endplate selection differences, the AI achieved robust agreement with human observers and standardized the measurement process. This study underscores DEEP:SPINE-AS-01® potential to enhance clinical workflow by auto-mating a traditionally time-intensive and error-prone process. Its accuracy, efficiency, and re-producibility suggest its value in routine scoliosis evaluation, reducing the burden on physicians and improving patient care. Future studies with larger datasets and diverse populations could further validate its clinical utility.