Multi-Class Flat Classification of Lung Diseases Utilizing Deep Learning

Abstract

Early diagnosis of diseases is very critical for recovery. However, this is not always feasible due to the limited available staff or expensive and inadequate tools as we have witnessed in the recent COVID-19 pandemic. Lung diseases are life-threatening, but fortunately, they can be detected from X-ray images, which are cost-effective approaches. However, they need experts who are sometimes unavailable. Thus, using cutting-edge technology to diagnose diseases automatically and fast is the key solution to saving people’s lives. In this research, deep learning techniques have been utilized to classify several lung diseases in a cost-saving, time-saving, and efficient manner. Examples of lung diseases studied in this research are COVID-19, Lung Opacity, Pneumonia, and Tuberculosis. Several pre-trained deep learning models have been employed for flat multi-class classification of these lung diseases instead of using binary classification to recognize one disease from normal cases, as most state-of-the-art studies carry out. The models’ performance has been evaluated on imbalanced data of X-ray images with various resolutions and types. Finally, multiple measurements metrics have been utilized to evaluate the performance. The best accuracy achieved in this research is 95.643%.