An Explainable Deep Learning Approach for Pneumonia Detection from Chest X-Ray with Comparative Evaluation of EfficientNet-B0 and DenseNet121
Abstract
Pneumonia is a critical respiratory illness that remains a significant source of morbidity and mortality worldwide. This again stresses the need for effective and efficient diagnostic support systems.” Chest X-ray imaging is an integral part of pneumonia diagnosis. Manual interpretation of X-ray images is a time-consuming process and may vary from person to person. This study proposes an approach for the automatic detection of pneumonia using deep learning techniques and chest X-ray images. A publicly available dataset consisting of 5,863 chest X-ray images is used for the study. Preprocessing techniques such as resizing, normalisation, and data augmentation are employed on the images to enhance the generalisation capability of the model. Two state-of-the-art architectures of convolutional neural networks, DenseNet121 and EfficientNet-B0, are used for the study. EfficientNet-B0 was shown to perform better in classification than DenseNet121 based on accuracy and reliability after the models were tested and assessed. Gradient-weighted Class Activation Map (Grad-CAM) was utilized to emphasize the pertinent regions of the chest X-ray images that the model concentrated on for the classification in order to enhance the interpretability of the suggested model. The findings of the experiment demonstrated that pneumonia may be detected using the suggested explainable deep learning framework.