Artificial Intelligence Computer-Aided Diagnosis to automatically predict the Pediatric Wrist Trauma using Medical X-ray Images

Pediatric wrist trauma (PWT) is a common injury that occurs in hospital emergency departments, with fractures being among the most frequent cases. The traditional diagnostic process for these injuries involves the collaboration of radiologists and surgeons. However, recent advancements in deep learning-based computer vision algorithms have shown potential in automating and expediting this diagnosis. In this paper, we propose an automatic end-to-end computer-aided diagnosis (CAD) system that can simultaneously detect and classify various types of pediatric wrist injuries, including bone anomalies, bone lesions, foreign bodies, fractures, metallic artifacts, periosteal reactions, pronator signs, soft tissue abnormalities, textual elements, and the axis. To train and evaluate the proposed CAD system, we utilize the public GRAZPEDWRI-DX dataset. The design of our CAD system involves careful steps such as medical X-ray data collection, data labeling, preprocessing, prediction AI model optimization, and model evaluation. The core component used for the detection and classification task in our system is YOLOv8, the advanced and state-of-the-art version of YOLO for object detection. Experimental results demonstrate that our proposed CAD system achieves promising evaluation results in terms of precision (77.80%), recall (54.60%), mAP@50 (59.10%), and mAP50@95 (37.20%). These encouraging evaluation results can serve as a future direction to provide practical solutions for improving the capabilities of the healthcare system in rapidly and accurately predicting pediatric wrist injuries.