A resource-efficient computer-aided diagnosis detector, G-YOLOv11, that adapts the YOLOv11 object detector for detecting fractures in pediatric wrist X-rays. It replaces standard convolutions with ghost convolution, which produces the same number of feature maps using fewer linear operations, sharply reducing model size and computation. On the GRAZPEDWRI-DX dataset the smallest variant trades a modest drop in mean average precision for a large reduction in model size, making real-time on-device fracture detection more practical in clinical settings.
This paper introduces G-YOLOv11, a lightweight computer-aided diagnosis detector for fractures in pediatric wrist X-rays, built on the YOLOv11 object detector. It replaces standard convolution with g…