E-CNN: ensembled CNN learning approach for pneumonia detection in chest X-ray images

Abstract

Pneumonia, a substantial global public health issue, is a pulmonary illness characterized by inflammation in the air sacs in either one or both lungs and can be triggered by several pathogens, such as bacteria, viruses, and fungi. In this regard, deep learning (DL) methods have demonstrated their effectiveness in classifying pneumonia which is a severe respiratory illness that specifically targets the lungs, particularly when analyzing chest X-ray (CXR) images. Therefore, significant efforts are being made to identify the source of the infection to assist clinicians in accurately diagnosing the etiology of the disease. The objective of this work is to investigate the efficacy of deep learning in the classification of pneumonia, utilizing transfer learning and the proposed weighted ensemble approach by utilizing a CXR dataset available publicly at the Kaggle repository comprising healthy and pneumonia cases caused by either viral or bacterial origins with a total of 5863 images. The preliminary results show our weighted ensemble approach stands out in performance achieving an accuracy of 98.40% significantly. The result demonstrates that current structures can accurately diagnose the sources of pneumonia and our deep CNN-based weighted ensemble learning approach could be a valuable tool in diagnosing upcoming unidentified diseases.

Topics & Concepts