X-Ray Image Enhancement Using Blind Denoising Neural Networks

X-ray imaging is a common medical imaging technology, which plays an important role in assisting doctors in diagnosis. However, the quality of X-ray images is often disturbed by noise in practical applications, which affects the diagnosis of the disease. Although several works have discussed the X-ray images denoising algorithms, their performance needs further improvement. In order to improve the quality of X-ray images, we propose a blind denoising algorithm based on convolutional neural network (X-BDCNN) with a more reasonable noise model. The noise model is designed according to the physical principle of the X-ray imaging, which can generate more realistic noisy X-ray images for training. X-BDCNN consists of a noise level estimation subnetwork and a non-blind denoising subnetwork. The noise level estimation subnetwork estimates the noise level of input noisy image so as to promote the performance of denoised image in the other subnetwork. Additionally, we add a SSIM loss function for X-BDCNN to further improve the quality of denoised images. The experiments under different noise levels demonstrate that our X-BDCNN has a superior performance in various evaluation metrics compared with existing denoising methods.