Comparative analysis of YOLO variants and EfficientNet for detecting bone fractures in X-ray images

Abstract

A bone fracture is a serious medical problem, and accurate and prompt diagnosis is crucial for optimal treatment. This study highlights the progress of automatic bone fracture detection using deep learning (DL) models. A dataset containing 17 different fracture classes was used to train and evaluate the models. The dataset had class imbalance and minor fracture detection challenges. Extensive preprocessing, including data augmentation and resizing, has been applied to solve these problems, which has helped to increase the robustness of the model. Seven state-of-the-art models—you only look once (YOLO)v8, YOLOv9, YOLOv10, YOLOv11, EfficientNetB0, DenseNet169 and ResNet50—are trained and evaluated. Precision, recall, F1-score, and mean average precision (mAP) were used to evaluate the performance of the models. Among all models, YOLOv11 leads the others by achieving the highest precision, mAP, and precision-recall balance. YOLOv11 adds architectural improvements such as a deep backbone network and hybrid feature fusion, which make the model more reliable in different types of fracture detection. It is capable of reducing false detections and maintaining stable memory usage consistency even under different imaging conditions. Overall, YOLOv11 showed promising results and highlighted the potential of AI-powered diagnostic tools to improve clinical processes and patient care. As future work, the application field of the model can be extended to larger medical imaging tasks, and it can be further refined for effective use in resource-limited environments.