Explainable Deep Learning Models on the Diagnosis of Pneumonia

Abstract

Pneumonia is a lung infection caused by pathogens such as bacteria and viruses. If the diagnosis is not timely and the treatment cannot be performed in time, it may threaten the life of the patient. Currently, the most common method for the diagnosis of pneumonia is to take chest X-rays and perform manual diagnosis. However, manual diagnosis is less efficient and highly subjective. In this paper, we present a deep learning approach for X-ray image identification of pneumonia and conducts the corresponding explainability. The essential idea is to use the existing pre-trained deep learning model and the Grad-CAM method to construct an explainable deep learning model of pneumonia X-ray image recognition. In the proposed deep learning model, ResNet50 and VGG16 models can achieve higher accuracy after data augmentation. The proposed deep learning model could be applied to the computer-aided diagnosis of pneumonia to achieve efficient detection and diagnosis of pneumonia.