Evaluating CNN and Deep Learning Models for Bone Fracture Detection: A Comparative Study of VGG19, ResNet-50, LeNet and DenseNet

Bone fracture detection is a critical topic in medical diagnostics due to the need for accurate and timely identification of fractures. Traditional methods, while reliable, can be time-consuming. Recent advancements in deep learning offer promising solutions for improving the accuracy and efficiency of fracture detection. This study evaluates the performance of various Convolutional Neural Network (CNN) and deep learning models in detecting bone fractures from medical imaging. Specifically, the comparison of four prominent models: VGG19, ResNet-50, LeNet, and DenseNet. These models are assessed based on their accuracy, sensitivity, specificity, and computational efficiency. The study implements the considered deep learning models on a publicly available bone fracture dataset from Kaggle, aiming to achieve precise and efficient detection results. The models, including VGG19, ResNet-50, LeNet-5, and DenseNet, are trained and tested on an organized set of X-ray images. Performance evaluation is based on the comparison of loss and accuracy across training, testing, and validation datasets. The results show that LeNet-5 and VGG19 demonstrate high detection accuracy, with stable results of 98.4% and 97.6%, respectively. This comparative analysis highlights the strengths and weaknesses of each model, offering valuable insights for their practical application in clinical environments.