An Image Dataset for Lung Disease Detection and Classification

Abstract

Lung diseases are part of several fatal illnesses. Although there are advances in the healthcare domain, some of them still top the list of worldwide mortal diseases. This paper analyzes how X-ray images, processed according to Artificial Intelligence, can be used to assist medical physicians and radiologists in their diagnosis by automatic detection and classification of lung diseases. In this study, we present a new image dataset that consists of 1071 chest X-ray images with two common lung pathologies, i.e. pneumonia and tuberculosis. In addition, with the help of medical specialists, we manually provided each image with disease bounding boxes. We used the proposed dataset to train Deep Learning based object detection models to demonstrate that thoracic pathologies can be automatically detected, classified and most importantly, localized. Our results are promising and show that the proposed dataset allows training accurate lung disease detection models.