Scoliosis can affect the patient’s physical appearance, cause pain, and cause difficulty breathing. With the continuous progress of medical technology and the continuous development of image processing technology, the accuracy of early detection and diagnosis of this disease has been improved. Therefore, this article aims to apply computer simulation technology to detect scoliosis images using models and improve the level of early intervention. This article mainly designs a detection model and generates images of scoliosis using computer simulation technology based on experimental and comparative methods, tests the performance of four models, compares the accuracy, recall, and F1 value of the selected Faster Region Convolutional Neural Networks (Faster R-CNN), analyzes the image generation results of the training, validation, and testing sets, and explores the capabilities of the three image generation techniques. The experimental results show that when the learning rate is 0.001, the Faster R-CNN model has the highest classification accuracy (0.92). Image transformation technology has strong robustness in generating scoliosis structures.