With advances in artificial intelligence, the use of computer-aided detection and diagnosis in clinical imaging is gaining traction. Typically, very large datasets are required to train machine-learning models, potentially limiting use of this technology when only small datasets are available. This study investigated whether pretraining of fracture detection models on large, existing datasets could improve the performance of the model when locating and classifying wrist fractures in a small X-ray image dataset. This concept is termed “transfer learning”. Firstly, three detection models, namely, the faster region-based convolutional neural network (faster R-CNN), you only look once version eight (YOLOv8), and RetinaNet, were pretrained using the large, freely available dataset, common objects in context (COCO) (330000 images). Secondly, these models were pretrained using an open-source wrist X-ray dataset called “Graz Paediatric Wrist Digital X-rays” (GRAZPEDWRI-DX) on a (1) fracture detection dataset (20327 images) and (2) fracture location and classification dataset (14390 images). An orthopaedic surgeon classified the small available dataset of 776 distal radius X-rays (Introduction
Method
As a part of the European Union BIOMED I study “Assessment of Bone Quality in Osteoporosis,” Sixty-nine second lumbar vertebral body specimens (L2) were obtained post mortem from 32 women and 37 men (age 24–92 years). Our initial remit was to study variations in density of the calcified tissues by quantitative backscattered electron imaging (BSE-SEM). To this end, the para-sagittal bone slices were embedded in PMMA and block surfaces micro-milled and carbon coated. Many samples were re-polished to remove the carbon coat and stained with iodine vapour to permit simultaneous BSE imaging of non-mineralised tissues - especially disc, annulus, cartilage and ligament - uncoated, at 50Pa chamber pressure. We have now studied most of these samples by 30-μm resolution high contrast resolution X-ray microtomography (XMT), typically 72 hours scanning time, thus giving exact correlation between high resolution BSE-SEM and XMT. The 3D XMT data sets were rendered using Drishti software to produce static and movie images for visualisation and edification. We have now selected a set of the female samples for reconstruction by 3D printing - taking as examples the youngest, post-menopausal, oldest, best, worst, and anterior and central compression fractures and anterior collapse with fusion to L3 - which will be attached to the poster display. The most porotic cases were also the most difficult to reconstruct. A surprising proportion of elderly samples showed excellent bone architecture, though with retention of fewer, but more massive, load-bearing trabeculae.
