PneuX-CNN: A Novel Architecture for Pneumonia Classification and Grad-Cam Explainability

Pneumonia is an infection that results in thickening of the air spaces in either of the lungs. It is extremely hazardous for elderly adults and infants. Therefore, it is imperative that we diagnose it at the earliest opportunity. In the conventional method of diagnosis, a patient undergoes a variety of tests, including a chest X-ray, blood test, and sputum analysis, which are subsequently evaluated by medical professionals. The evaluation of these tests and their results typically necessitates a significant amount of time. In certain instances, the disease can be fatal, particularly for small children, if the physicians or diagnosis centers are unable to detect it at an early stage. Therefore, it is essential to design a system that can easily and quickly identify pneumonia at a low cost. To identify pneumonia from chest X-ray images, numerous researchers implemented a deep learning (DL) approach. In this study, an automated enhanced model is proposed which can predict the presence of pneumonia in a person by automatically extracting features from images. The proposed model obtained an exactness of 98.20% by utilizing the image data augmentation technique and image normalization, which is more accurate than many other models. Lastly, the Grad-CAM interpretation gives us a better understanding of the proposed model’s output.