A NOVEL DEEP LEARNING APPROACH FOR EARLY DIAGNOSIS OF PNEUMONIA

Abstract

Pneumonia is a serious respiratory infection that causes the air sacs in the lungs to fill with fluid, leading to difficulty in breathing. It poses a significant public health challenge, particularly in countries like Pakistan, where limited access to healthcare facilities and the high cost of treatment contribute to high prevalence rates, especially among children. Early detection is essential to prevent the rapid spread and severe outcomes of the disease. This study aims to develop an automated pneumonia detection system using deep learning techniques. A pre-trained Convolutional Neural Network (CNN) and the ResNet-50 model were employed to classify chest X-ray images as either normal or pneumonia cases. The system achieved an accuracy of 97% using ResNet-50 and 94% using CNN. A comparative evaluation was also conducted using the YOLO-v5 model, which achieved an accuracy of 84%. The dataset included 6,083 chest X-ray images, with 5,086 used for training and validation and 223 images collected from a local hospital for testing. The system demonstrated strong performance in distinguishing between normal, viral, and bacterial pneumonia cases. To enhance accessibility, especially in rural areas, the model was deployed through a user-friendly web application called “Pneu Scan,” enabling real-time diagnosis based on uploaded chest X-ray images.