Deep Learning for Pneumonia Diagnosis Using CXR Images

Pneumonia is a life-threatening infection that affects the lungs caused by bacterial or virus attacks. It fills the air sacs of the lungs with pus or fluid. The factors that cause pneumonia are cystic fibrosis, asthma, sickle cell disease, diabetes, heart failure, and a weak immune system. Moreover, pneumonia is usually diagnosed by an expert radiologist by examining CXR images, however, it is a difficult and time-consuming task when images are unclear. In addition, after recent advancements in image processing, automatic detection of pneumonia is now achievable; nevertheless, existing research still faces overfitting and efficiency issues. This research proposed the Convolutional Neural Network (CNN) based approach for classifying normal and pneumonia-affected images. A publicly available dataset named Chest X-ray images comprised of 5,856 CXR images, has been utilized for training and testing the methodology. The proposed model consists of four stages: resizing, pre-processing, data augmentation, and classification of the images. Initially, images are resized into 448*448 pixels’ resolution and then pre-processed to remove noise from images. After that, various data augmentation approaches were applied to solve the overfitting problem. Additionally, a modified CNN model is used for the classification process. The experimental performance of the model has shown 97.5% of accuracy, 98.3% of recall, 97.4% of precision, and 98.9% of F1 score. Hence, the proposed framework can serve as an efficient application in pneumonia diagnosis and assist doctors in decision-making.