Bone Fracture Identification in Upper Limbs of Children Using Deep Learning Techniques

Purpose: The study employed CNN based deep learning algorithms including VGG16 and VGG19 architectures, to differentiate between fractured and normal bones in the upper limbs of children using X-ray images. Methods: Radiographic scans, consisting of 100 images (50 fractured and 50 normal) of upper limb bones were collected. The study conducted a comparison between deep learning models such as VGG16 and VGG19 architectures, and a computer aided diagnostic (CAD) software tool to evaluate their prediction performance in detecting bone fracture abnormalities. Results: In the proposed study, VGG19 classifier exhibited exceptional performance compared to other classifiers, surpassing them in accuracy, precision, and sensitivity. The dataset analysis indicated that the VGG19 model achieved an accuracy of 98.3%, while the VGG16 model recorded an accuracy of 92.5%. Conclusions: Deep learning has proven to be an efficient method for categorizing and identifying fractures and non-fractures in radiographs of children's upper limb anomalies.