Gradient regularized convolutional neural networks for low-dose CT image enhancement.

摘要: The potential risks of x-ray to patients have transferred the public's attention from normal dose CT (NDCT) low-dose (LDCT). However, simply lowering radiation system will significantly degrade quality images such as noise and artifacts, which compromises diagnostic performance. Hence, various methods been proposed solve this problem over past decades. Although these achieved impressive results, they also suffer a drawback smoothing image details after denoising, makes it difficult for clinical diagnosis treatment. To address issue, paper introduces novel gradient regularization method LDCT enhancement. Rather than common only consider pixel-wise gray value loss in reconstruction procedure, we take into consideration preserve details. By combining convolutional neural network (CNN) framework, regularized (GRCNN) is enhance has promising performance our experiments both visually quantitatively.