Chien Chun-Tse ; Ju Rui-Yang ; Chou Kuang-Yi ; Xieerke Enkaer ; Chiang Jen-Shiun

Abstract

Wrist trauma and even fractures occur frequently in daily life, particularlyamong children who account for a significant proportion of fracture cases.Before performing surgery, surgeons often request patients to undergo X-rayimaging first and prepare for it based on the analysis of the radiologist. Withthe development of neural networks, You Only Look Once (YOLO) series modelshave been widely used in fracture detection as computer-assisted diagnosis(CAD). In 2023, Ultralytics presented the latest version of the YOLO models,which has been employed for detecting fractures across various parts of thebody. Attention mechanism is one of the hottest methods to improve the modelperformance. This research work proposes YOLOv8-AM, which incorporates theattention mechanism into the original YOLOv8 architecture. Specifically, werespectively employ four attention modules, Convolutional Block AttentionModule (CBAM), Global Attention Mechanism (GAM), Efficient Channel Attention(ECA), and Shuffle Attention (SA), to design the improved models and train themon GRAZPEDWRI-DX dataset. Experimental results demonstrate that the meanAverage Precision at IoU 50 (mAP 50) of the YOLOv8-AM model based on ResBlock +CBAM (ResCBAM) increased from 63.6% to 65.8%, which achieves thestate-of-the-art (SOTA) performance. Conversely, YOLOv8-AM model incorporatingGAM obtains the mAP 50 value of 64.2%, which is not a satisfactory enhancement.Therefore, we combine ResBlock and GAM, introducing ResGAM to design anothernew YOLOv8-AM model, whose mAP 50 value is increased to 65.0%. Theimplementation code for this study is available on GitHub athttps://github.com/RuiyangJu/Fracture_Detection_Improved_YOLOv8.