Automated Abnormality Classification of Chest Radiographs Using MobileNetV2

Abstract

Chest X-ray is one of the most common screening and diagnostic radiological examinations for the detection of many lung diseases. Undoubtedly, evaluation of patient data and expert decisions are the most important factors in diagnosis. However, expert systems for classification and different artificial intelligence techniques also help experts a lot. Deep Learning, which has been widely used recently, is an advanced machine learning technique with many intangible layers that communicate with each other. In this study, chest disease was diagnosed using MobileNetV2, a popular deep learning network. X-ray image quality was tried to be improved by applying a three-steps pre-process including crop, histogram equalization and contrast-limited adaptive histogram equalization to data sets. The best result performance was given using ROC curve. Chest disease was detected by AC 89.95% and AUC 92.60 % using pre-processed ChestX-ray14 data sets.