<![CDATA[Fracture detection in pediatric wrist radiographs is challenging due to incomplete skeletal ossification, small bone structures, and subtle hairline fractures that are frequently missed in clinical practice, while growth plate radiolucency often mimics fracture appearance. This study evaluates YOLOv11n, a lightweight deep learning architecture with Spatial Pyramid Feature Fusion (SPFF) modules optimized for small-object detection, for automated pediatric wrist fracture identification. The model was trained and validated on the GRAZPEDWRI-DX benchmark dataset comprising 20,327 pediatric wrist radiographs (14,269 training, 4,048 validation, 2,010 test images) using transfer learning and conservative augmentation strategies. YOLOv11n achieved mAP@50 of 0.936 on validation and 0.945 on test sets, with precision of 0.905–0.918 and recall of 0.869–0.871, demonstrating improved accuracy compared to previous YOLOv8 implementations (mAP@50 ≈ 0.92) with 40–60% faster inference. End-to-end processing averaged 3.8 ms per image on NVIDIA Tesla T4 hardware, supporting real-time clinical applications. The mAP@50-95 of approximately 0.56 indicates reduced localization accuracy under stricter IoU criteria, primarily for hairline fractures. Error analysis revealed that 62% of false negatives were non-displaced hairline fractures, while 58% of false positives occurred near growth plate regions. YOLOv11n provides favorable balance between diagnostic accuracy and computational efficiency for pediatric fracture detection. However, prospective multi-institutional validation, integration of multi-view fusion strategies, and incorporation of age-specific anatomical priors are necessary before clinical deployment to enhance detection of subtle fracture presentations and reduce growth plate misclassifications.]]>
Copyrights © 2025
