Boring Survey Based Fracture Detection (BSFD) for Fragility Fracture of the Pelvis in CT Images

Becoming a super-aging society, fragility fracture of the pelvis (FFP) with osteoporosis is a common disease. However, it is hard to find FFP in CT images because of the numerous images and tiny appearance of FFP. This paper proposes a novel method, which is called boring survey based fracture detection (BSFD), to automatically detect FFP in 3D CT images. FFP appears as a tiny crack of the bone that is hard to find by surface observation. Thus, BSFD surveys the FFP by boring from the pelvic surface. It first segments the pelvic bone region from CT images. At every surface point, it bores a small quadrangular prism whose center is the point of interest. The prism is composed of voxels with CT values and is feed to a 3D convolutional neural network which predicts a probability of fracture. The proposed method was evaluated by using 110 elderly subjects with pelvic fractures. The AUC was 0.84 for training subjects and 0.77 for evaluation subjects. In addition, 3D visualization of fracture probability superimposed on the pelvic bone surface is provided for qualitative evaluation and supporting FFP diagnosis by physicians. The boring survey approach will be effective to detect FFP because the bored volume extracts the unique characteristics of tiny fractures in CT images.