3D reconstruction of X-ray skeletal images

X-ray image enhancement technology effectively improves the contrast between pathological regions and normal tissues, thereby enhancing image segmentation performance. High-quality segmentation results provide reliable sources for three-dimensional reconstruction. This study proposes a three-stage processing framework: (1) X-ray super-resolution enhancement based on the Cbc-SwinIR model (Coordinate-based Convolution - Image Restoration Using Swin Transformer); (2) the application of the MAXIM model (Multi-Axis MLP for Image Processing) to remove artifacts from X-ray images, followed by segmentation of target regions using theSAMmodel (Segment and Recognize Anything at Any Granularity); (3) monocular 3D reconstruction of X-ray images utilizing the One-2-3-45 model. Experimental results demonstrate outstanding performance in X-ray image enhancement and segmentation. A comprehensive series of tests was conducted on the X-ray image dataset collected in this study. By reconstructing image data corresponding to the angles of the original 2D images from multiple viewpoints of the generated 3D X-ray images, we obtained an average SSIM of 0.86, an average PSNR of 35.70, and an average MSE of 0.0155. The generation time for each 3D image was approximately 45 seconds, demonstrating both the effectiveness and real-time capability of the proposed algorithm.