Comparative Study of CNN Variants for Pneumonia Detection in Medical Imaging

As per the 2021 Global Burden of Disease report, 2.1 million people died of pneumonia in that year. Pneumonia is a major threat to children's lives. According to 2019 reports, around 45 million young children fall sick with pneumonia, leading to over 700,000 deaths, most of which were from the under 5 age group. Pneumonia is an illness of the lungs that triggers inflammation in the alveoli, the air sacs that facilitate breathing, where the alveoli get filled with a fluid or pus which leads to difficulty in breathing. The common symptoms are fever and chills, cough, shortness of breath, etc. It can also affect both the lungs or a single lung. In remote areas due to the lack of hospitals and resources and due to high cost many people lose their lives. The objective of the project is to create a machine learning system capable of predicting pneumonia through the analysis of chest X-ray images. The project aims to enhance early detection and reduce diagnostic errors. Its focus is on deep learning techniques, i.e., Convolutional Neural Networks (CNNs), for prediction, which has an accuracy of 92%. Various machine learning techniques, such as CNN, DenseNet, ResNet, InceptionNet, and VGG-16, were implemented and assessed. After analyzing their performance, CNN was found to deliver the most accurate results.