Detection of Pneumonia Utilising Deep learning-based Feature Extraction

Abstract

Pneumonia is a potentially fatal infectious disease that impacts one or both lungs in humans, primarily caused by the bacterium Streptococcus pneumoniae. According to the World Health Organisation (WHO), pneumonia is responsible for one in three deaths in India. Chest X-rays utilised for pneumonia diagnosis require assessment by specialist radiologists. Consequently, creating an automated method for pneumonia detection will facilitate timely treatment of the condition, especially in distant regions. Convolutional Neural Networks (CNNs) have garnered significant attention for illness classification due to the efficacy of deep learning algorithms in analysing medical imagery. Moreover, characteristics acquired by pre-trained CNN models on extensive datasets are highly beneficial for image classification tasks. This study evaluates the efficacy of pre-trained CNN models employed as feature extractors in conjunction with various classifiers for the classification of abnormal and normal chest X-rays. We analytically ascertain the best CNN model for the objective. The statistical results obtained indicate that pretrained CNN models, when utilised with supervised classifier methods, are highly advantageous for analysing chest X-ray pictures, particularly for the detection of pneumonia.