Fig. 3

Illustration of the process for extracting targeted bone images from segmentation outputs, and a comparison of two multi-crop augmentation methods. (Top) Seven bone image patches, each outlined by a square in the same color as the corresponding segmentation mask overlaid on the hand X-ray image. (Bottom) Each of the seven image segments undergoes two different multi-crop augmentation strategies: the conventional method (represented by a gray dashed line) and our extended method (indicated by a solid black line). A global view matching the color of the input bone patch and five local views are displayed side by side.