Classical total variation (TV) based iterative reconstruction algorithms assume that the signal is piecewise smooth, which causes reconstruction results to suffer from the over-smoothing effect. To address this problem, this work presents a novel computed tomography (CT) reconstruction method for the few-view problem called the group-sparsity regularization-based simultaneous algebraic reconstruction technique (GSR-SART). Group-based sparse representation, which utilizes the concept of a group as the basic unit of sparse representation instead of a patch, is introduced as the image domain prior regularization term to eliminate the over-smoothing effect. By grouping the nonlocal patches into different clusters with similarity measured by Euclidean distance, the sparsity and nonlocal similarity in a single image are simultaneously explored. The split Bregman iteration algorithm is applied to obtain the numerical scheme. Experimental results demonstrate that our method both qualitatively and quantitatively outperforms several existing reconstruction methods, including filtered back projection, expectation maximization, SART, and TV-based projections onto convex sets.

Few-View CT Reconstruction with Group-Sparsity Regularization

/pdf/plug-and-play-gradient-based-denoisers-applied-to-ct-image-uengujem.pdf

Plug-and-Play gradient-based denoisers applied to CT image enhancement

Abstract The rice quality is directly affected by the natural environment in which it is grown. It is common to pass off poor quality as good quality rice sold in rice markets. Therefore, an effective rice quality detection method should be proposed. This work proposes a deep learning method combined with an electronic nose (e-nose) to realize the classification of rice origin, and then realize the detection of rice quality. Firstly, channel-spatial collaborative attention (CSCA) is proposed to adaptively focus on key features of gas information. Secondly, three shallow convolutional neural network (CNN) structures are introduced to achieve gas information classification. Finally, CSCA and three CNN structures are combined to construct a classification network for rice gas recognition. The results show that in the comparison of multi-structure models, LeNet5-CSCA has the best classification performance, with a classification accuracy of 97.70%, a precision of 97.45%, a recall of 97.67%, and a F1-score of 97.56%. 

A deep learning method combined with electronic nose to identify the rice origin

Abstract Conventional reconstruction techniques, such as filtered back projection (FBP) and iterative reconstruction (IR), which have been utilised widely in the image reconstruction process of computed tomography (CT) are not suitable in the case of low-dose CT applications, because of the unsatisfying quality of the reconstructed image and inefficient reconstruction time. Therefore, as the demand for CT radiation dose reduction continues to increase, the use of artificial intelligence (AI) in image reconstruction has become a trend that attracts more and more attention. This systematic review examined various deep learning methods to determine their characteristics, availability, intended use and expected outputs concerning low-dose CT image reconstruction. Utilising the methodology of Kitchenham and Charter, we performed a systematic search of the literature from 2016 to 2021 in Springer, Science Direct, arXiv, PubMed, ACM, IEEE, and Scopus. This review showed that algorithms using deep learning technology are superior to traditional IR methods in noise suppression, artifact reduction and structure preservation, in terms of improving the image quality of low-dose reconstructed images. In conclusion, we provided an overview of the use of deep learning approaches in low-dose CT image reconstruction together with their benefits, limitations, and opportunities for improvement. 

https://link.springer.com/content/pdf/10.1007/s40747-022-00724-7.pdf

The use of deep learning methods in low-dose computed tomography image reconstruction: a systematic review

Contrast and depth resolution of breast lessions in a digital breast tomosynthesis system

Few-view computed tomography (CT) is able to provide the internal information of an object for diagnosis, while reducing radiation dose and decreasing side effects. In spite of these benefits, few-view CT images with conventional reconstruction algorithms are suffer from structural distortion and artifacts due to insufficient projections. In this study, a deep learning-based network was proposed for reconstructing few-view CT images and improving image quality. The proposed network consisted of fully connected, convolution and deconvolution layers, and the network training was implemented to directly transform few-view projections into reconstructed images without expert reconstruction knowledge. The quality of the output images obtained from the trained network was compared with those reconstructed by using the filtered back-projection (FBP) and simultaneous algebraic reconstruction technique (SART) algorithms. The results showed that the proposed network is able to improve the quantitative accuracy and noise property of reconstructed images, and the proposed network is more suitable for few-view CT reconstruction than the other algorithms. In conclusion, the proposed network can be potentially used to improve few-view CT image quality and considered as an alternative for few-view CT reconstruction. The source code is available on a GitHub repository at https://github.com/ YIMDOBIN/CT_reconstruction_via_DL .

Deep learning-based image reconstruction for few-view computed tomography

Computed tomography (CT) has been commonly used for providing medical images. However, CT scans have a potential risk of side-effect due to high radiation dose. In order to overcome this limitation, low-dose CT imaging techniques have been developed by using novel scanners, reconstruction algorithms and adjusted scanning protocols. In spite of these strategies, the conventional low-dose CT imaging techniques are hard to compromise between radiation dose and image quality, leading to noise, blur and artifacts in CT images. These drawbacks of the conventional techniques degrade diagnostic accuracy. In this study, we proposed a deep convolutional neural network (CNN)-based low-dose CT imaging technique for simultaneously reducing noise and blur at once. The proposed method, called denoising and deblurring CNN (DnDbCNN), consisted of two networks, and each network was separately trained for reducing noise and blur in CT images. For combining the trained networks, a convolutional layer was added between the two trained networks, and the combined network was re-trained to update the weights of each layer. Denoising convolutional neural network (DnCNN) and super-resolution convolutional neural network (SRCNN) were also implemented for comparison. We evaluated the performance of the DnDbCNN in terms of mean-square error (MSE), peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), coefficient-of-variation (COV) and modulation transfer function (MTF). The results showed that the DnDbCNN improved the noise property, spatial resolution and quantitative accuracy of CT images, and the performance of the DnDbCNN was superior to the conventional methods. As a result, the DnDbCNN can simultaneously reduce noise and blur by combining two networks through three training steps. Therefore, the DnDbCNN has a potential for improving CT image quality with low radiation dose and enables precise diagnosis.

Dobin Yim

Papers

Deep learning-based image reconstruction for few-view computed tomography

A deep convolutional neural network for simultaneous denoising and deblurring in computed tomography

Development of a truncation artifact reduction method in stationary inverse-geometry X-ray laminography for non-destructive testing

Analysis of Image Quality According to Imaging Parameters in Digital Tomosynthesis

Image Quality Evaluation in Computed Tomography Using Super-resolution Convolutional Neural Network