Abstract
Since absorption of X-ray radiation has the possibility of inducing cancerous, genetic and other diseases to patients, researches usually attempt to reduce the radiation dose. However, reduction of the radiation dose associated with CT scans will unavoidably increase the severity of noise and artifacts, which can seriously affect diagnostic confidence. Due to the outstanding performance of deep neural networks in image processing, in this paper, we proposed a Stacked Competitive Network (SCN) approach to noise reduction, which stacks several successive Competitive Blocks (CB). The carefully handcrafted design of the competitive blocks was inspired by the idea of multi-scale processing and improvement the network’s capacity. Qualitative and quantitative evaluations demonstrate the competitive performance of the proposed method in noise suppression, structural preservation, and lesion detection.
Highlights
Data Availability Statement: The resource code of this work has been released with 0https://github. com/Wenchao-Du/Stacked Competitive Network (SCN)-for-Image-Denoising0
Since the convolutional neural network (CNN)-based approaches are immune to the impact of the statistical distribution of the artifacts and noise, we model the noise reduction problem for low-dose CT (LDCT) as follows
In order to evaluate the performance of the proposed SCN, two typical slices, which were from thorax and abdomen, were selected
Summary
Data Availability Statement: The resource code of this work has been released with 0https://github. com/Wenchao-Du/SCN-for-Image-Denoising0. Projection space filtering [2] directly operates on raw projection data or log-transformed sinogram before FBP is applied This kind of method has low computational cost, their results may suffer from structure distortion due to the lack of well definition image edges in projection domain. Speaking, post-image processing methods can be applied directly on low-dose CT (LDCT) images and are more convenient to be combined into current CT systems. To prevent from discard of meaningful structural details, most neural network models for image restoration, which is considered a low-level task, had limited the depth of network This property is different from high-level tasks in computer vision, e.g. classification or detection [26][27], in which max-pooling operation is extensively used to capture high-level features.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have