Classification of lung diseases from X-ray images using deep learning

Abstract

The lung disease, due to COVID-19 for example, has caused devastation around the world. Even in the most developed nations, the growing number of cases has overwhelmed healthcare facilities. Radiographic imaging is still the most convenient screening method for lung diseases. A certified radiologist interprets the chest X-ray image according to their experience level. As such, the interpretations might vary for different radiologists based on the observed characteristics and due to possibility of human error. To counter this problem, an automated lung disease classification system using chest X-ray was proposed. The classification was achieved by using deep learning approach because artificial intelligence has been proven to help reduce human error in medical applications. In this project, five deep learning architectures namely ResNet18, ResNet50, ResNet101, Alexnet, and VGG16 architectures were selected for transfer learning and classification of lung diseases. The lung X-ray images were classified into five output classes, namely COVID-19, pneumonia, tuberculosis, nodule or normal lungs. Images from multiple public datasets were acquired to be used as train set and test set for this automated lung disease classification model. The five deep learning models were successfully tested, and the highest accuracy was 96.3%, achieved with the Alexnet architecture.