Chest X ray image enhancement using deep contrast diffusion learning

Abstract

The chest x-ray (CXR) is a diagnostic imaging tool that aids in the early detection and diagnosis of lung abnormalities. Due to scattering radiation, the CXR would have poor contrast, and the diagnosis would be difficult. Although many methods exist, deep learning (DL)-based CXR image enhancement remains difficult due to the amount of contrast that needs to be enhanced and the locations where acceptable contrast must be extracted. In order to improve CXR images, a contrast diffusion network is introduced in this paper. The input image is initially placed through a multi-level contrast-limited adaptive histogram Equalization (CLAHE) process, from which the necessary contrast is extracted and sent into the convolutional neural network (CNN)-based residual learning network together with low contrast CXR. To create the enhanced CXR images, the learned contrast features were diffused over the input image. The amount of contrast to be diffused is determined by multiple levels of CLAHE. Various metrics are used to evaluate the enhanced image's quality. Additionally, the enhanced images are submitted to computer-assisted diagnosis, which improves overall classification efficiency. All of the results are based on the Shenzhen, COVID-CXR, and PadChest datasets.