Detection of hidden pediatric elbow fractures in X-ray images based on deep learning

Abstract

BackgroundPediatric elbow fractures pose a common diagnostic challenge in the field of pediatric orthopedics. Accurate diagnosis of these fractures through X-ray images is crucial in clinical practice, often requiring highly specialized medical imaging experts. In recent years, the development of deep learning technology has provided a new approach to address this issue. This study aims to explore the potential of using deep convolutional neural network (DCNN) technology, specifically the DenseNet-201 model, for diagnosing pediatric elbow fractures from X-ray images. MethodsThis study utilized X-ray images from 1370 pediatric patients, obtained from multiple healthcare centers. The images were carefully selected based on specific inclusion and exclusion criteria, ensuring a diverse representation of elbow fractures. Each image underwent preprocessing, including normalization and resizing, to standardize the input for the DenseNet-201 model. The DensenNet-201 model was trained using a split of 70% for training and 30% for validation. Key training parameters included a learning rate of 0.001, 50 epochs, and a batch size of 32. Advanced techniques like dropout and data augmentation were employed to enhance the model's ability to generalize and prevent overfitting. The performance of the models was evaluated using metrics such as accuracy, sensitivity, specificity, and area under the curve (AUC). ResultsThe research findings demonstrate the outstanding performance of the DenseNet-201 model across all datasets. It achieved an accuracy of 94.1% and an AUC of 98.7% on the training dataset, indicating its robust discriminatory ability in accurately diagnosing elbow fractures. Furthermore, the model exhibited excellent sensitivity (93.2%) and specificity (94.8%) in the training dataset, indicating its ability to accurately detect positive and negative cases of elbow fractures. ConclusionThe results of this study emphasize the potential of the DenseNet-201 model as a diagnostic tool for pediatric elbow fractures. Accurate and timely diagnosis of pediatric elbow fractures is crucial for effective treatment and patient prognosis. The exceptional performance of the DenseNet-201 model, particularly in terms of sensitivity and specificity, suggests that it can assist clinical practitioners in confidently diagnosing these fractures, thereby reducing misdiagnosis rates and improving the quality of patient care. It should be noted that multiple models were compared, with DenseNet-201 demonstrating the most prominent performance.