Detection of pneumonia from pediatric lung x-ray images with convolutional neural network

In this study, it has been researched on the detection of pneumonia disease, that seriously threats children and might cause deaths. Lung x-ray displays have an important role in detection of this disease. Because those devices which are used to gain this diaplays are more common than any other display devices and spreads lower radiation. The analysis of lung x-ray displays are made by radiologists. However, some adverse conditions such as vague image, similarity of sympthoms with other diseases and intense work schedule may make it difficult to analyze x-ray displays. In this study, it has been developed Convolutional neural network models which can classify lung x-ray displays correctly (with pneumonia) and help the experts. Two different approach has been used while modelling convolutional neural network. The first is modelling the network via determining all layers at the beginning originally, and the second is modelling the network using transfer learning methodology. Convolutional neural networks generally result better when the related data set size is bigger. In this study, a ready dataset composed of previously displayed 5840 x-ray images of children under age 5. Although this dataset is relatively small for the suggested method, It caused the memorization of the model in the training of the network. After memorization, it has been observed that correlation difference between the training and test accuracy grows. In order to reduce this differance, data increase and regulation methods were used and the training of the network is significantly better. In the first model that was originally designed, %93.43 accuracy rate was gained; on the other hand %97.12 accuracy rate was gained in transfer learning model. High success rate reached in the results of the study shows that suggested methods can help experts to analyze lung x-ray displays. Besides, suggested methodology may help quick and accurate diagnosis in epidemic periods (such as COVID-19/a kind of pneumonia).