Clinical Decision Support Systems for Pneumonia Diagnosis Using Gradient-Weighted Class Activation Mapping and Convolutional Neural Networks

In recent years, Deep Learning (DL) has gained great achievements in medicine. More specifically, DL techniques have had unprecedented success when applied to Chest X-Ray (CXR) images for disease diagnosis. Numerous scientists have attempted to develop efficient image-based diagnosis methods using DL algorithms. Their proposed methods can yield very reasonable performance on prediction tasks, but it is very hard to interpret the generated output from such deep learning algorithms. In this study, we propose a Convolutional Neural Network (CNN) architecture combining Gradient-weighted Class Activation Mapping (Grad-CAM) algorithm to discriminate between pneumonia patients and healthy controls as well as provide the explanations for the generated results by the proposed CNN architecture. The explanations include regions of interest that can be signs for the considered disease. As shown from the results, the proposed method has achieved a promising performance and it is expected to help the radiologists and doctors in the diagnosis process.