Pneumonia detection in X-ray images using convolutional neural networks

Through this paper it is demonstrated the application of artificial intelligence in the field of medicine, namely that of radiology. Pneumonia is an infectious disease that is widespread throughout the world. It is associated with disease, high mortality, and also high costs for the healthcare system. Diagnosing this disease quickly is very important to save lives. Using Deep Learning, respectively the Convolutional Neural Networks algorithm, a model from scratch has been built to serve as a supportive CAD system (Computer-aided Detection System) for radiologists in the detection of pneumonia in chest X-Ray images. Three other models have been built using the transfer learning technique, ready-made architectures built on very large amounts of data such as ResNet from Microsoft, VGGNet from Oxford and Inception from Google. Each model is documented, evaluated and compared. Different parameters and techniques were used in each model to achieve great classification performance. The data to train the model were taken from the set of the chest X-Ray images made public for research purposes. The results obtained from the built classification models are good, and can be useful to many. The best classification accuracy was achieved using the transfer learning technique where Inception was used as the base model and an accuracy of 92.7% was obtained. Whereas from the model built from scratch an accuracy of 89.5% has been obtained