A Novel Method for Pneumonia Diagnosis from Chest X-Ray Images Using Deep Residual Learning with Separable Convolutional Networks

Abstract

Pneumonia is an infection that inflames the air sacs in one or both lungs. The air sacs may fill with fluid or pus, causing cough with phlegm or pus, fever, chills, and difficulty in breathing. Pneumonia can be caused by a variety of organisms such as bacteria, viruses, and fungi. Apart from causing difficulty in respiration, pneumonia can cause other complications such as bacteremia, lung abscess, and pleural effusion, among countless others. This paper presents a novel automated method for efficient and accurate pneumonia diagnosis from chest X-ray(CXR) images. Also, this model utilizes an efficient technique for noise reduction using bilateral filtering for edge preservation and optimum enhancement of the images using contrast-limited adaptive histogram equalization(CLAHE) to aid the detection of pneumonia clouds in the CXR images. The proposed model explores the benefits of deep residual learning along with separable convolution algorithm to achieve a classification accuracy of $$98.82\%$$ and AUROC score of 0.99726 for diagnosing the disease. For cross-validation of the model, gradient-weighted class activation map(Grad-CAM) and saliency map visualization are used as a measure to verify the performance of the model along with localization of the affected regions in the CXR images.