An Effective Cost-Sensitive Learning Approach for Detection of COVID-19 with Lung Diseases
摘要
Using a chest X-ray to locate COVID-19 has become highly challenging due to issues with class balance. This study proposes using convolutional neural networks (CNNs) to improve COVID-19 identification performance and address the class imbalance problem. The proposed method uses cost-sensitive learning with a CNN model specifically designed to address the problem. Chest X-ray images of individuals with COVID-19 and lung diseases were used in this study. Most images were typical, and the dataset was somewhat erratic. However, there were a few pictures of pneumonia and COVID-19. This study uses three different approaches to address the uneven distribution of information and prevent getting stuck: (i) data resampling, (ii) algorithm modifications, and (iii) ensemble methods. Based on experimental results, the proposed method performs better than prior state-of-the-art techniques with a higher detection rate of COVID-19 and better performance metrics like accuracy, recall, and F1-score. The proposed method shows promise with a sensitivity of 88.89% and an accuracy of 91.67%. The CNN model's feature maps determine which areas of the chest X-ray images help identify COVID-19. Cost-sensitive learning has been proposed as the best solution to address the class imbalance problem in COVID-19 identification through medical imaging. The recommended approach proved effective in the study report for locating COVID-19 in chest X-rays of patients with lung conditions that lead to class disparity. To validate its therapeutic utility and generalizability, more investigation is necessary.