A NOVEL APPROACH FOR CHEST DISEASE DETECTION USING DEEP LEARNING & FEDERATED LEARNING MODELS

Abstract

Healthcare industry serves as a major sector of economy of every country. It is a well- defined practice that started several years ago. As the time passed the improvement in healthcare sector has touched the new edges and of course technology has always played a very important role in it. Each and every phase of the healthcare development has different issue with them. When it comes to the prehistoric era, when medications were not yet created, healing with herbs and other natural resources was a lengthy process that may take years to complete. . Now-a-days with advancement of machines and technology although the time involved in treatment has decreased but the problem arises to store data and records of several patients, records, treatments and many more. This research work focuses on the detection of several chest diseases including lymphoma disease. We have employed DTL and FL models to make the prediction procedure more accessible and faster such as VGG-16, MobileNet V2, ResNet50, DenseNet161, Inception V3, and VGG-19. As we realize that chest diseases are so regular now-a-days, it's basic to successfully predict and analyze them. The study's dataset of 112,120 chest x-ray pictures was analysed. Images from 30,805 people with a total of fourteen different types of chest disorders—including atelectasis, consolidation, infiltration, and pneumothorax—as well as a class dubbed "No findings" if the condition was undetected—were included in the study. The research's proposed system was developed to predict 14 different kinds of chest conditions, and a comparison between federated learning and deep learning models was made. The best model was the VGG-16, which had an accuracy of 0.81 and a recall rate of 0.90. As a result, the F1 score of the VGG-16 is 0.85, higher than the F1 scores attained with alternate transfer learning procedures. Through federated transfer learning, VGG-19 achieved a maximum accuracy of 96.71%. Consequently, the classification report said that the VGG-16 model was the most effective transfer-learning model for accurately identifying chest disease. Likewise, Lymphoma histopathological diagnosis is a difficult task that needs specialized knowledge or centralized evaluation. The proposed study describes a system that can predict lymphoma cancer with reasonable accuracy using sample images taken by various pathologists at different locations. A dataset titled malignant lymphoma classification dataset, which contains 5400 images, was used to analyze this study. Chronic lymphatic leukemia (CLL), follicular lymphoma (FL), and mantle cell lymphoma (MCL) are the three sorts of lymphoma. Notwithstanding, just two of them were concentrated on in this review: CLL and FL. The suggested system in the research is used to forecast two forms of lymphoma cancer, and a comparison of deep learning and federated learning models had been made. Four threshold values—0.05, 0.1, 0.15, and 0.2—were used for each class in the classification process, with the parameters precision, recall, and F1 score being taken into consideration. When the models were evaluated, it was discovered that the VGG-19 model that was suggested had the best accuracy, most minimal loss, and root mean square error by 95.8%, 0.055, and 0.234 when using the unified learning procedure, and the best precision, least loss, and root mean square error by 97.5%, 0.155, and 0.393 when using the deep exchange learning strategy. Consequently, the classification report states that the VGG-19 model is the most effective deep and federated transfer-learning model for lymphoma cancer classification.