Deep Learning Reconstruction Method Research Articles

Improvement of depiction of the intracranial arteries on brain CT angiography using deep learning reconstruction.

To evaluate the ability of a commercialized deep learning reconstruction technique to depict intracranial vessels on the brain computed tomography angiography and compare the image quality with filtered-back-projection and hybrid iterative reconstruction in terms of objective and subjective measures. Forty-three patients underwent brain computed tomography angiography, and images were reconstructed using three algorithms: filtered-back-projection, hybrid iterative reconstruction, and deep learning reconstruction. The image noise, computed tomography attenuation value, signal-to-noise ratio, and contrast-to-noise ratio were measured in the bilateral cavernous segment of the internal carotid artery, vertebral artery, basilar apex, horizontal segment of the middle cerebral artery and used for the objective assessment of the image quality among the three different reconstructions. The subjective image quality score was significantly higher for the deep learning reconstruction than hybrid iterative reconstruction and filtered-back-projection images. The deep learning reconstruction markedly improved the reduction of blooming artifacts in surgical clips and coiled aneurysms. The deep learning reconstruction method generally improves the image quality of brain computed tomography angiography in terms of objective measurement and subjective grading compared with filtered-back-projection and hybrid iterative reconstruction. Especially, deep learning reconstruction is deemed advantageous for better depiction of small vessels compared to filtered-back projection and hybrid iterative reconstruction.

Image quality assessment of pediatric chest and abdomen CT by deep learning reconstruction

BackgroundEfforts to reduce the radiation dose have continued steadily, with new reconstruction techniques. Recently, image denoising algorithms using artificial neural networks, termed deep learning reconstruction (DLR), have been applied to CT image reconstruction to overcome the drawbacks of iterative reconstruction (IR). The purpose of our study was to compare the objective and subjective image quality of DLR and IR on pediatric abdomen and chest CT images.MethodsThis retrospective study included pediatric body CT images from February 2020 to October 2020, performed on 51 patients (34 boys and 17 girls; age 1–18 years). Non-contrast chest CT (n = 16), contrast-enhanced chest CT (n = 12), and contrast-enhanced abdomen CT (n = 23) images were included. Standard 50% adaptive statistical iterative reconstruction V (ASIR-V) images were compared to images with 100% ASIR-V and DLR at medium and high strengths. Attenuation, noise, contrast to noise ratio (CNR), and signal to noise (SNR) measurements were performed. Overall image quality, artifacts, and noise were subjectively assessed by two radiologists using a four-point scale (superior, average, suboptimal, and unacceptable). A phantom scan was performed including the dose range of the clinical images used in our study, and the noise power spectrum (NPS) was calculated. Quantitative and qualitative parameters were compared using repeated-measures analysis of variance (ANOVA) with Bonferroni correction and Wilcoxon signed-rank tests.ResultsDLR had better CNR and SNR than 50% ASIR-V in both pediatric chest and abdomen CT images. When compared with 50% ASIR-V, high strength DLR was associated with noise reduction in non-contrast chest CT (33.0%), contrast-enhanced chest CT (39.6%), and contrast-enhanced abdomen CT (38.7%) with increases in CNR at 149.1%, 105.8%, and 53.1% respectively. The subjective assessment of overall image quality and the noise was also better on DLR images (p < 0.001). However, there was no significant difference in artifacts between reconstruction methods. From NPS analysis, DLR methods showed a pattern of reducing the magnitude of noise while maintaining the texture.ConclusionCompared with 50% ASIR-V, DLR improved pediatric body CT images with significant noise reduction. However, artifacts were not improved by DLR, regardless of strength.

Accurate and robust sparse-view angle CT image reconstruction using deep learning and prior image constrained compressed sensing (DL-PICCS).

Sparse-view CT image reconstruction problems encountered in dynamic CT acquisitions are technically challenging. Recently, many deep learning strategies have been proposed to reconstruct CT images from sparse-view angle acquisitions showing promising results. However, two fundamental problems with these deep learning reconstruction methods remain to be addressed: (1) limited reconstruction accuracy for individual patients and (2) limited generalizability for patient statisticalcohorts. The purpose of this work is to address the previously mentioned challenges in current deep learningmethods. A method that combines a deep learning strategy with prior image constrained compressed sensing (PICCS) was developed to address these two problems. In this method, the sparse-view CT data were reconstructed by the conventional filtered backprojection (FBP) method first, and then processed by the trained deep neural network to eliminate streaking artifacts. The outputs of the deep learning architecture were then used as the needed prior image in PICCS to reconstruct the image. If the noise level from the PICCS reconstruction is not satisfactory, another light duty deep neural network can then be used to reduce noise level. Both extensive numerical simulation data and human subject data have been used to quantitatively and qualitatively assess the performance of the proposed DL-PICCS method in terms of reconstruction accuracy andgeneralizability. Extensive evaluation studies have demonstrated that: (1) quantitative reconstruction accuracy of DL-PICCS for individual patient is improved when it is compared with the deep learning methods and CS-based methods; (2) the false-positive lesion-like structures and false negative missing anatomical structures in the deep learning approaches can be effectively eliminated in the DL-PICCS reconstructed images; and (3) DL-PICCS enables a deep learning scheme to relax its working conditions to enhance itsgeneralizability. DL-PICCS offers a promising opportunity to achieve personalized reconstruction with improved reconstruction accuracy and enhanced generalizability.

Validation of a deep learning reconstruction framework for 3D delayed myocardial enhancement imaging

Abstract Funding Acknowledgements Type of funding sources: None. Introduction Myocardial delayed enhancement (MDE) MRI plays an important role in the identification of several cardiac conditions, both ischemic and non-ischemic (e.g. myocarditis, IDC, amyloidosis). 3D imaging offers increased resolution, full heart coverage and better depiction of complex pathologies, but its image quality is limited by long acquisition times. Deep learning (DL) models enable advanced reconstruction algorithms that yield regularized images in practical computation times. In this study we evaluate a novel 3D-DL reconstruction to overcome the trade-off between reconstructed quality and acquisition time on MDE data. Methods A group of 14 subjects referred for CMR (5 F / 9 M, 59 ± 11 y.o., 78 ± 13 kg) were scanned with a 3D MDE sequence prototype: SPGR with IR preparation, fat &amp; spatial saturation, respiratory navigator, ARC 2x, FOV 40x40cm, ST 1.4-2.4mm, matrix 280²-320², FA 20deg, BW 62.5 kHz, TE 2.1 ± 0.1ms, TI based on a CINE IR scout. All were retrospectively reconstructed using a 3D DL algorithm, trained on a database of over 700 datasets to reconstruct high-quality images with adjustable noise reduction. The images were compared with standard 3D Cartesian reconstruction by two experienced cardiologists, to identify alterations in morphology or contrast distribution. Noise was estimated using the intensity standard deviation on a blood pool ROI. Feature preservation was estimated using the structural similarity index (SSI). Results The new method improved perceived image quality without loss of structural information or resolution (fig 1). Quantitative analysis (fig 2) confirmed these results: The average coefficient of variation in the blood was 0.08 ± 0.02 in the reference and 0.05 ± 0.02 with the new method; Given a target image noise level, DL reconstruction yielded up to 10% better SSI, compared to anisotropic filtering. The clinical review didn’t reveal diagnostically significant alterations of structure or uptake pattern. A perceived reduction of sharpness was initially reported but individual examination of landmarks (e.g. pulmonary and coronary arteries) confirmed that no relevant features were being lost with the new reconstruction. Discussion The 3D MDE images obtained with DL reconstruction improved the trade-off between image noise -estimated by the blood pool intensity deviation- and feature preservation -estimated by SSI-. Consistent improvement of image quality without morphological alterations of diagnostic relevance indicates that the new method can be considered for clinical practice. The next step in the validation process will require testing the robustness over a large set of cases with heterogeneous acquisition settings. Conclusion We presented the preliminary evaluation of a deep learning reconstruction method with 3D myocardial delayed enhancement data. The results show systematic improvement of overall image quality without loss of relevant diagnostic information. Abstract Figure.

A partial convolution-based deep-learning network for seismic data regularization1

Spatial undersampling is a common problem in actual seismic data due to limitations in seismic survey environments, which can be satisfactorily solved by data regularization. The convolution-based deep-learning reconstruction methods require fewer assumptions than the conventional reconstruction methods (e.g., Curvelet-domain and F-X domain data regularization methods). However, the traditional convolution methods are not suitable for the large percentages of missing data. In this study, we propose an improved partial convolution-based (PConv-based) deep-learning network to reconstruct the missing data, which is evolved from the conventional convolution-based (CConv-based) method. The U-net is used as deep learning network to analyze both PConv-based method and CConv-based method. The PConv-based method adopts a hierarchical, regional-learning mechanism to dynamically update the constrained convolution results for the sample matrix. Hence, the problem of poor amplitude preservation in the data reconstruction has been addressed when multiple consecutive traces are missing. The influence of data loss ratio on reconstruction algorithm is also discussed in this study. The numerical test demonstrates that the trained network is able to process a sample dataset with 50% data lost and largely eliminate the noises in the frequency-wavenumber domain caused by the missing data. This proposed method is further evaluated by actual data, and the results are better than those obtained from the Curvelet-domain method. Moreover, the dataset reconstructed by the PConv-based deep-learning network has a great agreement with the original dataset in terms of amplitude.

LANTERN: Learn analysis transform network for dynamic magnetic resonance imaging

&lt;p style='text-indent:20px;'&gt;This paper proposes to learn analysis transform network for dynamic magnetic resonance imaging (LANTERN). Integrating the strength of CS-MRI and deep learning, the proposed framework is highlighted in three components: (ⅰ) The spatial and temporal domains are sparsely constrained by adaptively trained convolutional filters; (ⅱ) We introduce an end-to-end framework to learn the parameters in LANTERN to solve the difficulty of parameter selection in traditional methods; (ⅲ) Compared to existing deep learning reconstruction methods, our experimental results show that our paper has encouraging capability in exploiting the spatial and temporal redundancy of dynamic MR images. We performed quantitative and qualitative analysis of cardiac reconstructions at different acceleration factors (&lt;inline-formula&gt;&lt;tex-math id="M1"&gt;\begin{document}$ 2 \times $\end{document}&lt;/tex-math&gt;&lt;/inline-formula&gt;-&lt;inline-formula&gt;&lt;tex-math id="M2"&gt;\begin{document}$ 11 \times $\end{document}&lt;/tex-math&gt;&lt;/inline-formula&gt;) with different undersampling patterns. In comparison with two state-of-the-art methods, experimental results show that our method achieved encouraging performances.&lt;/p&gt;

Deep learning-enhanced fluorescence microscopy via degeneration decoupling.

Deep learning-based reconstruction has emerged as an effective tool in fluorescence microscopy, with the potential to resolve diffraction-limited structures. However, most deep-learning reconstruction methods employed an end-to-end strategy, which ignored physical laws in the imaging process and made the preparation of training data highly challenging as well. In this study, we have proposed a novel deconvolution algorithm based on an imaging model, deep-learning priors and the alternating direction method of multipliers. This scheme decouples the reconstruction into two separate sub-problems, one for deblurring and one for restraining noise and artifacts. As a result of the decoupling, we are able to introduce deep-learning image priors and a variance stabilizing transform against targeted image degeneration due to the low photon budget. Deep-learning priors are learned from the general image dataset, in which biological images do not have to be involved, and are more powerful than hand-designed ones. Moreover, the use of the imaging model ensures high fidelity and generalization. Experiments on various kinds of measurement data show that the proposed algorithm outperforms existing state-of-the-art deconvolution algorithms in resolution enhancement and generalization.

MRI reconstruction using deep Bayesian estimation.

To develop a deep learning-based Bayesian estimation for MRI reconstruction. We modeled the MRI reconstruction problem with Bayes's theorem, following the recently proposed PixelCNN++ method. The image reconstruction from incomplete k-space measurement was obtained by maximizing the posterior possibility. A generative network was utilized as the image prior, which was computationally tractable, and the k-space data fidelity was enforced by using an equality constraint. The stochastic backpropagation was utilized to calculate the descent gradient in the process of maximum a posterior, and a projected subgradient method was used to impose the equality constraint. In contrast to the other deep learning reconstruction methods, the proposed one used the likelihood of prior as the training loss and the objective function in reconstruction to improve the image quality. The proposed method showed an improved performance in preserving image details and reducing aliasing artifacts, compared with GRAPPA, -ESPRiT, model-based deep learning architecture for inverse problems (MODL), and variational network (VN), last two were state-of-the-art deep learning reconstruction methods. The proposed method generally achieved more than 3 dB peak signal-to-noise ratio improvement for compressed sensing and parallel imaging reconstructions compared with the other methods. The Bayesian estimation significantly improved the reconstruction performance, compared with the conventional -sparsity prior in compressed sensing reconstruction tasks. More importantly, the proposed reconstruction framework can be generalized for most MRI reconstruction scenarios.