A study on deep learning-based classification for Pneumonia detection

In this paper, we investigate the various aspects of methodologies in deep learning-based pneumonia classification using chest x-ray images. As widely known, selecting appropriate hyper-parameters is essential for increasing the classification performance in convolution neural networks(CNN). We experiment with various hyper-parameters, including the number of layers, optimizer, learning rate, and momentum factor for diagnosing pneumonia using CNN. In addition, we test different CNN models and augmentation methods for chest x-ray diagnosing. Experimental results show that the proposed non-rigid transform based on augmentation increases classification accuracy by up to 5%.