Deep learning for childhood pneumonia detection from chest X-ray images

The popularity of studies to automatically detect diseases is increasing day by day. Recently, many researchers have proven that the use of Deep Learning (DL) exhibits better detection performance and easier classification process. Therefore, the number of DL-based diagnostic research articles continues to increase. In this study, deep learning models are recommended for the early detection of pneumonia, which is one of the respiratory tract diseases. Childhood pneumonia is one of the important causes of child mortality and accurate detection has a critical role in this regard. The proposed network models are trained on chest X-ray images of individuals with pneumonia and healthy individuals. Thus, the system helps in the early diagnosis of Pneumonia disease. These models propose the two-class classification model. Before the training of the models, certain preprocessing steps were applied to the data set and data augmentation was carried out with the SMOTE method. With this method, the number of training data sets belonging to both classes is equalized. Thus, the tendency to the more numerous classes in education is reduced and overfitting is prevented. The training of the network was carried out with CNN, RNN and LSTM algorithms from deep learning algorithms. Trained models detected pneumonia with CNN 97.23%, RNN 89.23% and LSTM 88.92% accuracy, outperforming the latest technology.