Automated Detection of Pneumonia from Digital Chest Radiographs Using Convolutional Neural Networks

Abstract

In the field of medical image processing, neural networks are frequently utilized for automating analysis and classification tasks. Diagnosing pneumonia from CT or X-ray scans is challenging due to subtle symptoms, highlighting the need for objective and automated diagnosis in medical imaging. Given the significant global burden of pneumonia-related mortality, addressing this challenge is paramount.

This chapter aims to propose a solution using advanced deep neural network architecture. The innovation lies in integrating residual blocks with down-sampling and convolutions in the convolutional segment of the network. The approach was trained and evaluated on labelled images from NIH datasets, focusing on anterior-posterior chest X-ray images of patients aged one to twelve from Third I Imaging, a diagnostic centre in India.

Our network's results demonstrate competitiveness with state-of-the-art models such as SVM, Decision Trees, Random Forests, and UNET, achieving 97.5% accuracy, 96.6% F Score, and 0.08 Error Rate. This research contributes to improving automated pneumonia diagnosis, addressing a critical need in medical imaging.