Hip fracture detection in x-ray images using deep learning

Abstract

Hip fractures, particularly in the femoral neck and intertrochanteric regions, pose serious health risks; if untreated, they can lead to severe complications and even mortality. Detecting and classifying these fractures in radiographs is challenging due to the subtle similarities in bone structure and positioning across images. This study addresses these challenges by developing an automated system using object detection models, specifically YOLO versions 8, 10, and 11, with various image augmentations to enhance detection accuracy. Model tuning was limited by the high similarity of details and structure in the images, so augmentations such as adjustments in color shades, brightness, rotation, shifting, and blending were applied to introduce diversity. Hardware constraints also restricted the use of resource-intensive models, such as using large batch sizes. Despite these challenges, newer model versions demonstrated improved performance, achieving a sensitivity rate of 98.1% for fracture detection and a mean average precision (mAP50) of 0.963, while another configuration achieved a mAP50-95 score of 0.724. These results suggest that enhancing the model’s backbone architecture could potentially yield even higher accuracy

Ähnliche Arbeiten