A Study on CNN Architectures for Chest X-Rays Multiclass Computer-Aided Diagnosis

Abstract

X-rays are the most commonly used medical images and are involved in all areas of healthcare because they are relatively inexpensive compared to other modalities and can provide sensitive results. The interpretation by the radiologist, however, can be challenging because it depends on his experience and a clear mind. There is also a lack of specialized physicians, mainly in the least developed areas, which increases the need for alternatives to X-ray analysis. Recent research shows that the development of Deep Learning based methods for chest X-rays analysis has the potential to replace the radiologists analysis in the future. However, most of the published DL algorithms were developed to classify a single disease. We propose an ensemble of Deep Neural Networks that can classify several classes. In this work, the network was used to classify five chest diseases: Atelectasis, Cardiomegaly, Consolidation, Edema, and Pleural Effusion. An AUC of 0.96 was achieved with the training data and 0.74 with the test data.