목적: This study aims to validate the clinical utility of AI-driven predictive software we developed for OSAS by analyzing anatomical data and airflow parameters, identifying optimal patient populations for surgical intervention, and determining specific airway regions that yield the most beneficial outcomes when performing surgery. 방법:A prospective, single-center study was conducted on 25 patients with moderate-to-severe OSAS (apnea-hypopnea index [AHI] ≥15 events/hour) who were intolerant to CPAP therapy. Patients underwent SP TORS TBR with expansion sphincter pharyngoplasty as a multilevel surgical approach. The study integrated computational fluid dynamics (CFD) and machine learning for airway analysis, employing a 3D UNet deep-learning model for automatic segmentation of upper airway structures from CT scans. A multivariate Gaussian process regression model was developed to predict airflow patterns rapidly, and a support vector machine classifier was used to distinguish between healthy subjects and OSAS patients. 결과:The mean patient age was 37.9 ± 11.9 years, with a mean BMI of 26.63 ± 3.52 kg/m². The median surgical duration was 116 minutes, with a median console time of 25 minutes. Correlation analysis revealed that tongue- epiglottis volume (AUC 1) and uvula-tongue volume (AUC 0.824) were significant predictors of surgical success. Significant predictors of surgical success included tongue-epiglottis volume (AUC = 1.0) and uvula-tongue volume (AUC = 0.824). Machine learning-based AHI classification showed high concordance with traditional polysomnography. Notably, most airway changes did not significantly correlate with AHI improvements, except for the anteroposterior width of the nasopharynx. 결론:AI-driven predictive modeling, combined with CFD and CT imaging, provides a novel framework for surgical planning and outcome prediction in OSAS. This approach identifies critical anatomical predictors of success, such as tongue-epiglottis and uvula-tongue volumes, supporting the advancement of personalized medicine in OSAS management. These findings suggest that AI-driven models could be integrated into preoperative workflows, enabling personalized surgical planning and potentially improving long-term patient outcomes in OSAS management. Additionally, CT-based assessments offer reliable adjunctive value alongside traditional diagnostic methods, enhancing the precision of therapeutic strategies.