Convolutional Neural Networks (CNNs) have the potential to significantly enhance the identification of humerus fractures on X-rays. However, deeper CNNs often lead to an increase in the size of parameters, which limits their use in devices with limited resources. Moreover, effective techniques are needed to measure humerus fracture detection at an early stage. In this paper, we propose a deep learning model called DenseResNet that improves the diagnosis of humeral fractures through the combination of DenseNet169 and the residual block. We tested the performance of our model with different loss functions, such as focal loss, Binary Cross Entropy (BCE), and Weighted Cross Entropy (WCE). We find that the DenseResNet (AUC of 0.920) outperforms the traditional DenseNet169 (AUC of 0.859). These enhancements indicate the potential of more efficient diagnostic tools in medical environments and suggest that similar modifications would be beneficial to other conventional backbone designs.
Azmi et al. (Mon,) studied this question.