Automated Pneumonia Detection using deep features in chest X-ray images

Abstract

Pneumonia is swelling of the lungs that is usually caused by an infection. This disease is considered as one of the most common reasons for US children to be hospitalized. According to American Thoracic Society (ATS), the cost of treating pneumonia cases in hospitals reached 9.5 billion dollar. The appropriate treatment and recovery process for this disease are linked to early diagnosis. In this work a novel method is proposed for detecting the pneumonia and help the radiologists in their decision making process. First, histogram equalization (HE) and Contrast Limited Adaptive Histogram Equalization (CLAHE) are calculated for chest X-ray images. Then, the images extracted are fed to a model consisting of two stream of Convolutional Neural Networks (CNN) that was trained on the Pneumonia Kermany dataset. Finally, several machine learning classifiers are employed to perform the detection process based on the deep features extracted. The proposed system achieves 97.86% in terms of accuracy on the Kermany dataset, which is satisfactory when compared to recently published works.