Accurate selection of lowest instrumented vertebra (LIV) tilt is essential to prevent distal adding-on in adolescent idiopathic scoliosis. We previously showed that achieving an intraoperative LIV tilt (β)

≥ preoperative Alpha Angle (α) from supine left side- bending radiographs reduced adding-on to 10%. This study aimed to develop an AI model to automate Alpha Angle calculation and provide an objective, patient-specific LIV tilt target. This study included 249 AIS patients (Lenke 1–6) undergoing posterior spinal fusion. Supine left side-bending radiographs were divided into training (n=199) and validation (n=50) sets. A YOLOv11-pose CNN was trained to identify T11–L4 vertebrae and the pelvis to calculate the Alpha Angle. Performance was validated against expert consensus using ICC, mean absolute error, clinical success rates (±3° and ±5°), and mean average precision. The AI system achieved robust technical performance on the validation set, with a bounding box mAP50 of 99.4% and pose estimation mAP50 of 97.6%. In clinical validation, the model demonstrated excellent reliability with an ICC of 0.931 (95% CI: 0.89–0.96). The Mean Absolute Error was 1.68°, with negligible systematic bias (-0.12°; p=0.78). The AI achieved a clinical success rate of 92.0% within a 5° threshold and 80.0% within a 3° threshold. AI- surgeon agreement rate at 3° was 80.0%, and at 5° was 94.0%. There was no significant difference in X-angle, Alpha angle, and adjusted Alpha angle between surgeon consensus and AI. For the Y-angle, the AI measured a significantly lower displacement (p-value<0.001). Stratified analysis showed robust accuracy from T11 to L1, though reliability decreased at the L2 level (ICC=0.39). This study introduces the first AI system to automate Alpha Angle calculation for AIS surgical planning, demonstrating expert- level reliability (ICC >0.93) and accuracy exceeding inter-surgeon variability. With 92% of measurements within 5° of expert consensus, it provides a rapid and consistent alternative to manual measurement, bridging complex modeling and practical clinical application.