Classification of Multiple Abnormalities for X-Ray Images with Deep-Learning-Based Framework

Abstract

Obtaining an early and accurate diagnosis of pneumonia is crucial for enhancing the cure rate and reducing the mortality rate. The interpretation of chest X-Rays is contingent upon the physician's experience, potentially resulting in bias and inaccurate diagnosis. A lot of previous studies focused on the improving the classification result as well as the quality of digital Chest X-Ray datasets. This study area has promise for future advancement, focusing on the classification of chest radiographs. To enhance the Chest Radiographs dataset of patients with lung disease, we proposed employing a multistage deep transfer learning technique that integrates an updated Generative Adversarial Network (GAN). A recommendation was given, and this was one of them. The proposal encompasses the execution of multi-stage transfer learning, to be achieved by the utilization of Inception V3 and Xception, respectively. The classification findings demonstrate a high level of competence, featuring an F1 score of 98.97%, an accuracy rate of 99.14%, a precision rate of 98.80%, a recall rate of 98.34%, and an accuracy rate of 99.14%. Future studies may focus on enhancing dataset quality and investigating methods to facilitate transfer learning by utilizing alternative combinations of learning models. Both of these focal areas present potential opportunities for future research. The ultimate objective of this research is to establish a society characterized by optimal health and well-being. This will be achieved by enhancing the healthcare system in alignment with Saudi Vision 2030.