Differential Evaluation Multi-scale U-NET (DEMSU-NET) Architecture for Classification of Lung Diseases from X-Ray Images

Worldwide, lung diseases are a common occurrence. It consists of pneumonia, asthma, TB, fibrosis, Chronic Obstructive Pulmonary Disease (COPD), and others. However the early detection of this disease is crucial. Thus several Machine Learning (ML) and image processing methods have been introduced for disease detection from images. Deep learning (DL) is an effective ML approach which integrates the procedure of supervised training by feature distribution and unsupervised training to shorten optimization. Lung disease diagnosis from Chest X-Ray (CXR) images has been extensively studied using the U-NET architecture. Differential Evaluation Multi-Scale U-NET (DEMSU-NET) Architecture, multi-scale feature maps are extracted from every convolutional of the U-NET encoder. Noisy or insufficient annotations may decrease the accuracy of U-NET model; it may be solved by auxiliary confidence maps. It takes place less emphasis on the limits of the provided target detection of lung disease. Differential Evaluation (DE) is implemented to adjust the background and foreground weights based on the population X-ray image. National Institutes of Health (NIH) chest X-ray images are gathered from the Kaggle repository to experiment the detection methods. Results of the proposed system and current methods are assessed using measures such as precision, recall, Fβ-score, and accuracy.