Application of deep learning in chest X-ray abnormality detection

Abstract

Lung diseases are the most common diseases worldwide, especially in Vietnam. Certain thoracic lung diseases can even lead to dangerous conditions for patients. X-ray are a useful imaging modality for detecting the abnormalities in the chest area. Furthermore, artificial intelligence can improve the detection of abnormalities in X-ray images, reduce misdiagnosis, close the knowledge gap between doctors, and alleviate the pressure on doctors. Therefore, this study aims to apply deep learning techniques to detect abnormalities in chest X-ray images and use data science and statistical approaches to improve the performance of the deep learning model. The data was explored and processed to obtain high quality data with optimal characteristics. We then applied data augmentation and optimization to the RetinaNet model with ResNet101 in a Feature Pyramid Network (FPN) backbone to achieve the best performance. Our model achieved mean average precision of 0.55 at a threshold of 0.5 (mAP@0.5) in a validation set, which included five diseases: aortic enlargement, cardiomegaly, interstitial lung disease, infiltration, and nodule/mass.