An Early Detection of Pneumonia in CXR Images using Deep Learning Techniques

Abstract

Pneumonia is a leading cause of death worldwide, and diagnosing other lung diseases, including lung cancer, cardiomegaly, atelectasis, etc., can be difficult. The most common technique for determining the presence of pneumonia is using chest X-ray imaging. However, analyzing a chest X-ray is a complex process that might result in significant subjective variation. In this research, one of the main goals is to figure out how to use deep learning (DL) to spot pneumonia on CXR. This study provides a CNN model for automatically detecting pneumonia in chest radiographs. This study has built an ensemble of three CNN models and used deep transfer learning (DTL) to deal with the data shortage. The methodology entails collecting a dataset of CXR images, which is then preprocessed, enhanced utilizing threshold, LNB feature extracted, data augmented to form a new data point and split into two datasets. In the end, CNN was utilized to teach about and categorize Pneumonia. With the suggested CNN approach, the greatest testing and training accuracy rates of 0.9888 and 0.9281 were obtained on the pneumonia detection (PD) dataset. These results are based on fusing the scores of four standard assessment metrics: precision, accuracy, recall, and f1-score.