Lung Diseases Classification from Chest X-Ray Images using Deep Learning

Abstract

Lung diseases are disorders in the lung that affect proper functioning of the breathing system. The top five prevalent lung diseases that are the leading cause of death include Chronic Obstructive Pulmonary Disease, Lung Cancer, Pneumonia, Tuberculosis, and Pneumothorax. Diagnosis of lung diseases is usually performed through visual inspection of chest X-ray images, especially in the developing world. This is time consuming, tedious, and subjected to inter and intra-observer variability which may lead to misdiagnosis. In this research, a method for automatic, accurate and reliable classification of the top five lung diseases from chest X-ray radiograph images is proposed using a deep learning approach. The data required for training, validation and testing the system was collected from online National Institute of Health chest X-ray14 dataset repository and the local data was acquired from Jimma University Medical Center radiology department. Deep learning approach based on Xception model was used for multi class classification task. All the images have been pre-processed prior to feeding to the model. The system has been developed using Python 3.7 programing language. A graphical user interface has been developed for ease of use and implementation. An accuracy, sensitivity and specificity of 97.3%, 97.2%, and 99.4%, respectively, have been achieved for multi-class classification using the proposed algorithm. The system takes only an average of 1 minute to provide the diagnosis result. The developed system will have a great impact in reducing the diagnosis errors imposed by the manual visual inspection method and can be used as a decision support system for physicians, especially those in low resource setting where both the expertise and the means is in scarce