Adolescent idiopathic scoliosis (AIS) is a spinal deformity occurring in adolescents, with a risk of progression into adulthood. Severe cases often require corrective surgery, typically through spinal fusion. However, AIS diagnosis exposes patients to repeated radiation from imaging modalities such as X-rays, and measurement of the Cobb angle is prone to inter- and intra-observer variability. Recent advances in artificial intelligence (AI) and deep learning, particularly generative adversarial networks (GANs), offer opportunities to predict surgical outcomes preoperatively, potentially improving surgical planning and patient care. We hypothesized that GAN-based models could accurately predict postoperative Cobb angles and generate postoperative radiographs from preoperative patient data, supporting their use as decision-support tools in AIS surgery. A retrospective dataset comprising clinical metadata from 153 AIS patients and 87 paired preoperative–postoperative radiographs were analyzed. Feature selection was performed using Variable Selection Using Random Forests (VSURF) to identify dominant clinical predictors. A conditional GAN (cGAN) was trained to predict postoperative Cobb angle from the selected predictors, while a Pix2Pix GAN was employed for paired image-to-image translation to generate postoperative radiographs. Model performance was evaluated using mean absolute error (MAE), root mean squared error (RMSE), coefficient of determination (R²), structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), Fréchet Inception Distance (FID), and Inception Score (IS). VSURF identified preoperative Cobb angle, Lenke curve type, and fusion levels as the most important predictors. The cGAN model predicted postoperative Cobb angle with a mean absolute error (MAE) of 4.11°, root mean squared error (RMSE) of 5.12°, and an R² of 0.8564 in predicting postoperative Cobb angle. The Pix2Pix model generated synthetic radiographs with an SSIM of 0.7475, PSNR of 24.38 dB, FID of 229.24, and IS of 1.04, demonstrating moderate structural similarity and perceptual realism compared to ground-truth images. GAN-based architectures are feasible for predicting AIS surgical outcomes from patient-specific data. These models have potential as precision decision-support tools to improve preoperative planning and patient care in spinal deformity surgery.