Deep Learning Based Noise Reduction for Brain MR Imaging: Tests on Phantoms and Healthy Volunteers

SDnDTI: Self-supervised deep learning-based denoising for diffusion tensor MRI

Diffusion tensor magnetic resonance imaging (DTI) is a widely adopted neuroimaging method for the in vivo mapping of brain tissue microstructure and white matter tracts. Nonetheless, the noise in the diffusion-weighted images (DWIs) decreases the accuracy and precision of DTI derived microstructural parameters and leads to prolonged acquisition time for achieving improved signal-to-noise ratio (SNR). Deep learning-based image denoising using convolutional neural networks (CNNs) has superior performance but often requires additional high-SNR data for supervising the training of CNNs, which reduces the feasibility of supervised learning-based denoising in practice. In this work, we develop a self-supervised deep learning-based method entitled "SDnDTI" for denoising DTI data, which does not require additional high-SNR data for training. Specifically, SDnDTI divides multi-directional DTI data into many subsets of six DWI volumes and transforms DWIs from each subset to along the same diffusion-encoding directions through the diffusion tensor model, generating multiple repetitions of DWIs with identical image contrasts but different noise observations. SDnDTI removes noise by first denoising each repetition of DWIs using a deep 3-dimensional CNN with the average of all repetitions with higher SNR as the training target, following the same approach as normal supervised learning based denoising methods, and then averaging CNN-denoised images for achieving higher SNR. The denoising efficacy of SDnDTI is demonstrated in terms of the similarity of output images and resultant DTI metrics compared to the ground truth generated using substantially more DWI volumes on two datasets with different spatial resolutions, b-values and numbers of input DWI volumes provided by the Human Connectome Project (HCP) and the Lifespan HCP in Aging. The SDnDTI results preserve image sharpness and textural details and substantially improve upon those from the raw data. The results of SDnDTI are comparable to those from supervised learning-based denoising and outperform those from state-of-the-art conventional denoising algorithms including BM4D, AONLM and MPPCA. By leveraging domain knowledge of diffusion MRI physics, SDnDTI makes it easier to use CNN-based denoising methods in practice and has the potential to benefit a wider range of research and clinical applications that require accelerated DTI acquisition and high-quality DTI data for mapping of tissue microstructure, fiber tracts and structural connectivity in the living human brain.

Whole-body MRI: detecting bone metastases from prostate cancer

Whole-body magnetic resonance imaging (WB-MRI) is currently used worldwide for detecting bone metastases from prostate cancer. The 5-year survival rate for prostate cancer is > 95%. However, an increase in survival time may increase the incidence of bone metastasis. Therefore, detecting bone metastases is of great clinical interest. Bone metastases are commonly located in the spine, pelvis, shoulder, and distal femur. Bone metastases from prostate cancer are well-known representatives of osteoblastic metastases. However, other types of bone metastases, such as mixed or inter-trabecular type, have also been detected using MRI. MRI does not involve radiation exposure and has good sensitivity and specificity for detecting bone metastases. WB-MRI has undergone gradual developments since the last century, and in 2004, Takahara et al., developed diffusion-weighted Imaging (DWI) with background body signal suppression (DWIBS). Since then, WB-MRI, including DWI, has continued to play an important role in detecting bone metastases and monitoring therapeutic effects. An imaging protocol that allows complete examination within approximately 30 min has been established. This review focuses on WB-MRI standardization and the automatic calculation of tumor total diffusion volume (tDV) and mean apparent diffusion coefficient (ADC) value. In the future, artificial intelligence (AI) will enable shorter imaging times and easier automatic segmentation.

Hybrid deep-learning-based denoising method for compressed sensing in pituitary MRI: comparison with the conventional wavelet-based denoising method

OBJECTIVES

This study aimed to evaluate the efficacy of a combined wavelet and deep-learning reconstruction (DLR) method for under-sampled pituitary MRI.

METHODS

This retrospective study included 28 consecutive patients who underwent under-sampled pituitary T2-weighted images (T2WI). Images were reconstructed using either the conventional wavelet denoising method (wavelet method) or the wavelet and DLR methods combined (hybrid DLR method) at five denoising levels. The signal-to-noise ratio (SNR) of the CSF, hypothalamic, and pituitary images and the contrast between structures were compared between the two image types. Noise quality, contrast, sharpness, artifacts, and overall image quality were evaluated by two board-certified radiologists. The quantitative and the qualitative analyses were performed with robust two-way repeated analyses of variance.

RESULTS

Using the hybrid DLR method, the SNR of the CSF progressively increased as denoising levels increased. By contrast, with the wavelet method, the SNR of the CSF, hypothalamus, and pituitary did not increase at higher denoising levels. There was a significant main effect of denoising methods (p < 0.001) and denoising levels (p < 0.001), and an interaction between denoising methods and denoising levels (p < 0.001). For all five qualitative scores, there was a significant main effect of denoising methods (p < 0.001) and an interaction between denoising methods and denoising levels (p < 0.001).

CONCLUSIONS

The hybrid DLR method can provide higher image quality for T2WI of the pituitary with compressed sensing (CS) than the wavelet method alone, especially at higher denoising levels.

KEY POINTS

• The signal-to-noise ratios of cerebrospinal fluid progressively increased with the hybrid DLR method, with an increase in the denoising level for cerebrospinal fluid in pituitary T2WI with CS. • The signal-to-noise ratios of cerebrospinal fluid using the conventional wavelet method did not increase at higher denoising levels. • All qualitative scores of hybrid deep-learning reconstructions at all denoising levels were higher than those for the wavelet denoising method.

Contrast-Enhanced Fluid-Attenuated Inversion Recovery in Neuroimaging: A Narrative Review on Clinical Applications and Technical Advances

While contrast-enhanced fluid-attenuated inversion recovery (FLAIR) has long been regarded as an adjunct sequence to evaluate leptomeningeal disease in addition to contrast-enhanced T1-weighted imaging, it is gradually being used for more diverse pathologies beyond leptomeningeal disease. Contrast-enhanced FLAIR is known to be highly sensitive to low concentrations of gadolinium within the fluid. Accordingly, recent research has suggested the potential utility of contrast-enhanced FLAIR in various kinds of disease, such as Meniere's disease, seizure, stroke, traumatic brain injury, and brain metastasis, in addition to being used for visualizing glymphatic dysfunction. However, its potential applications have been reported sporadically in an unorganized manner. Furthermore, the exact mechanism for its superior sensitivity to low concentrations of gadolinium has not been fully understood. Rapidly developing magnetic resonance technology and unoptimized parameters for FLAIR may challenge its accurate application in clinical practice. This review provides the fundamental mechanism of contrast-enhanced FLAIR, systematically describes its current and potential clinical application, and elaborates on technical considerations for its optimization. LEVEL OF EVIDENCE: 3 TECHNICAL EFFICACY STAGE: 5.

Content-Noise Complementary Learning for Medical Image Denoising

Medical imaging denoising faces great challenges, yet is in great demand. With its distinctive characteristics, medical imaging denoising in the image domain requires innovative deep learning strategies. In this study, we propose a simple yet effective strategy, the content-noise complementary learning (CNCL) strategy, in which two deep learning predictors are used to learn the respective content and noise of the image dataset complementarily. A medical image denoising pipeline based on the CNCL strategy is presented, and is implemented as a generative adversarial network, where various representative networks (including U-Net, DnCNN, and SRDenseNet) are investigated as the predictors. The performance of these implemented models has been validated on medical imaging datasets including CT, MR, and PET. The results show that this strategy outperforms state-of-the-art denoising algorithms in terms of visual quality and quantitative metrics, and the strategy demonstrates a robust generalization capability. These findings validate that this simple yet effective strategy demonstrates promising potential for medical image denoising tasks, which could exert a clinical impact in the future. Code is available at: https://github.com/gengmufeng/CNCL-denoising.

Deep Learning Reconstruction of Diffusion-weighted MRI Improves Image Quality for Prostatic Imaging

Background Deep learning reconstruction (DLR) may improve image quality. However, its impact on diffusion-weighted imaging (DWI) of the prostate has yet to be assessed. Purpose To determine whether DLR can improve image quality of diffusion-weighted MRI at b values ranging from 1000 sec/mm² to 5000 sec/mm² in patients with prostate cancer. Materials and Methods In this retrospective study, images of the prostate obtained at DWI with a b value of 0 sec/mm², DWI with a b value of 1000 sec/mm² (DWI₁₀₀₀), DWI with a b value of 3000 sec/mm² (DWI₃₀₀₀), and DWI with a b value of 5000 sec/mm² (DWI₅₀₀₀) from consecutive patients with biopsy-proven cancer from January to June 2020 were reconstructed with and without DLR. Image quality was assessed using signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) from region-of-interest analysis and qualitatively assessed using a five-point visual scoring system (1 [very poor] to 5 [excellent]) for each high-b-value DWI sequence with and without DLR. The SNR, CNR, and visual score for DWI with and without DLR were compared with the paired t test and the Wilcoxon signed rank test with Bonferroni correction, respectively. Apparent diffusion coefficients (ADCs) from DWI with and without DLR were also compared with the paired t test with Bonferroni correction. Results A total of 60 patients (mean age, 67 years; age range, 49-79 years) were analyzed. DWI with DLR showed significantly higher SNRs and CNRs than DWI without DLR (P < .001); for example, with DWI₁₀₀₀ the mean SNR was 38.7 ± 0.6 versus 17.8 ± 0.6, respectively (P < .001), and the mean CNR was 18.4 ± 5.6 versus 7.4 ± 5.6, respectively (P < .001). DWI with DLR also demonstrated higher qualitative image quality than DWI without DLR (mean score: 4.8 ± 0.4 vs 4.0 ± 0.7, respectively, with DWI₁₀₀₀ [P = .001], 3.8 ± 0.7 vs 3.0 ± 0.8 with DWI₃₀₀₀ [P = .002], and 3.1 ± 0.8 vs 2.0 ± 0.9 with DWI₅₀₀₀ [P < .001]). ADCs derived with and without DLR did not differ substantially (P > .99). Conclusion Deep learning reconstruction improves the image quality of diffusion-weighted MRI scans of prostate cancer with no impact on apparent diffusion coefficient quantitation with a 3.0-T MRI system. © RSNA, 2022 Online supplemental material is available for this article. See also the editorial by Turkbey in this issue.

[Improvement of Motion Artifacts in Brain MRI Using Deep Learning by Simulation Training Data]

PURPOSE

To test whether deep learning can be used to effectively reduce artifacts in MR images of the brain.

METHODS

In this study, a large set of images with and without motion artifacts is needed for training. It is difficult to collect training data from clinical images because it requires a lot of effort and time. We have created motion artifact images of the brain by computer simulation. As an experimental study, we obtained original images for deep learning from 20 volunteers. These original images were used to create various images of different artifacts by computer simulation and these were used the input images for deep learning. The same method was used to create test images and these images were used to compare the structural similarity (SSIM) index and peak signal-to-noise ratio (PSNR) between the input images and output images using the three denoising methods. The network models used were U-shaped fully convolutional network (U-Net), denoising convolutional neural network (DnCNN) and wide inference network and 5 layers Residual learning and batch normalization (Win5RB).

RESULTS

U-Net was the most effective model for reducing motion artifacts. The SSIM and PSNR were 0.978 and 32.5 dB.

CONCLUSION

This is an effective method to reduce artifacts without degrading the image quality of brain MRI images.

Deep Learning in Large and Multi-Site Structural Brain MR Imaging Datasets

Large, multi-site, heterogeneous brain imaging datasets are increasingly required for the training, validation, and testing of advanced deep learning (DL)-based automated tools, including structural magnetic resonance (MR) image-based diagnostic and treatment monitoring approaches. When assembling a number of smaller datasets to form a larger dataset, understanding the underlying variability between different acquisition and processing protocols across the aggregated dataset (termed “batch effects”) is critical. The presence of variation in the training dataset is important as it more closely reflects the true underlying data distribution and, thus, may enhance the overall generalizability of the tool. However, the impact of batch effects must be carefully evaluated in order to avoid undesirable effects that, for example, may reduce performance measures. Batch effects can result from many sources, including differences in acquisition equipment, imaging technique and parameters, as well as applied processing methodologies. Their impact, both beneficial and adversarial, must be considered when developing tools to ensure that their outputs are related to the proposed clinical or research question (i.e., actual disease-related or pathological changes) and are not simply due to the peculiarities of underlying batch effects in the aggregated dataset. We reviewed applications of DL in structural brain MR imaging that aggregated images from neuroimaging datasets, typically acquired at multiple sites. We examined datasets containing both healthy control participants and patients that were acquired using varying acquisition protocols. First, we discussed issues around Data Access and enumerated the key characteristics of some commonly used publicly available brain datasets. Then we reviewed methods for correcting batch effects by exploring the two main classes of approaches: Data Harmonization that uses data standardization, quality control protocols or other similar algorithms and procedures to explicitly understand and minimize unwanted batch effects; and Domain Adaptation that develops DL tools that implicitly handle the batch effects by using approaches to achieve reliable and robust results. In this narrative review, we highlighted the advantages and disadvantages of both classes of DL approaches, and described key challenges to be addressed in future studies.

Clinical validity of non-contrast-enhanced VI-RADS: prospective study using 3-T MRI with high-gradient magnetic field

OBJECTIVES

To develop a modified Vesical Imaging Reporting and Data System (VI-RADS) without dynamic contrast-enhanced imaging (DCEI), termed "non-contrast-enhanced VI-RADS (NCE-VI-RADS)", and to assess the additive impact of denoising deep learning reconstruction (dDLR) on NCE-VI-RADS.

METHODS

From January 2019 through December 2020, 163 participants who underwent high-gradient 3-T MRI of the bladder were prospectively enrolled. In total, 108 participants with pathologically confirmed bladder cancer by transurethral resection were analyzed. Tumors were evaluated based on VI-RADS (scores 1-5) by two readers independently: an experienced radiologist (reader 1) and a senior radiology resident (reader 2). Conventional VI-RADS assessment included all three imaging types (T2-weighted imaging [T2WI], diffusion-weighted imaging [DWI], and dynamic contrast-enhanced imaging [DCEI]). Also evaluated were NCE-VI-RADS comprising only non-contrast-enhanced imaging types (T2WI and DWI), and "NCE-VI-RADS with dDLR" comprising T2WI processed with dDLR and DWI. All systems were assessed using receiver-operating characteristic curve analysis and simple and/or weighted κ statistics.

RESULTS

Muscle invasion was identified in 23/108 participants (21%). Area under the curve (AUC) values for diagnosing muscle invasion were as follows: conventional VI-RADS, 0.94 and 0.91; NCE-VI-RADS, 0.93 and 0.91; and "NCE-VI-RADS with dDLR", 0.96 and 0.93, for readers 1 and 2, respectively. Simple κ statistics indicated substantial agreement for NCE-VI-RADS and almost perfect agreement for conventional VI-RADS and "NCE-VI-RADS with dDLR" between the two readers.

CONCLUSION

NCE-VI-RADS achieved predictive accuracy for muscle invasion comparable to that of conventional VI-RADS. Additional use of dDLR improved the diagnostic accuracy of NCE-VI-RADS.

KEY POINTS

• Non-contrast-enhanced Vesical Imaging Reporting and Data System (NCE-VI-RADS) was developed to avoid risk related to gadolinium-based contrast agent administration. • NCE-VI-RADS had predictive accuracy for muscle invasion comparable to that of conventional VI-RADS. • The additional use of denoising deep learning reconstruction (dDLR) might further improve the diagnostic accuracy of NCE-VI-RADS.

Visualization of the saccule and utricle with non-contrast-enhanced FLAIR sequences

OBJECTIVES

3D-fluid attenuation inversion recovery (FLAIR) collected 4 h after intravenous gadolinium injection can delineate the perilymphatic space (PLS) from the endolymphatic space (ELS) to capture endolymphatic hydrops, the pathological counterpart of Ménière's disease. We aimed to optimize visualization of such inner ear internal anatomy using 3D-FLAIR without injection.

METHODS

3D-FLAIR signal from different fluid compartments such as PLS and ELS was first simulated. Then, twenty-two healthy subjects were scanned at 3.0-T MRI with non-injected 3D-FLAIR using variable T2 preparations (T2Preps) (OFF, 200, 400, and 600 ms) and variable inversion times (TIs) (from 224 to 5000 ms) and different resolutions (1.0 × 1.0 × 1.5, 0.6 × 0.6 × 0.8, and 0.6 × 0.6 × 0.6 mm³). The relative contrast between PLS and ELS and the visibility of the saccule and utricle were assessed. Additionally, non-injected 3D-FLAIR with the optimal setting was tested in a Ménière patient and compared with gadolinium-injected 3D-FLAIR.

RESULTS

The PLS and ELS were differentiated when T2Prep was used but not without. The relative contrast was larger with T2Prep at 400 ms than at 200 or 600 ms (0.72 ± 0.22 vs. 0.44 ± 0.11, p = 0.019; and 0.72 ± 0.22 vs. 0.46 ± 0.28, p = 0.034, respectively). The saccule and utricle were best delineated in 87. % cases with T2Prep = 400 and TI = 2100 ms at the highest resolution. Visualization of the saccule and utricle in the optimized non-injected 3D-FLAIR was similar to conventional injected 3D-FLAIR in a patient.

CONCLUSIONS

Combining a specific T2Prep and TI in non-injected 3D-FLAIR could separate PLS and ELS and even the saccule and utricle, paving the way toward future application to diagnose Ménière's disease.

KEY POINTS

• MRI can capture the internal anatomy of inner ear without injection of contrast media. • Specific parameters consisting of a T2 preparation of 400 ms and an inversion time of 2100 ms must be used to visualize the saccule and utricle on non-injected 3D-FLAIR.

Machine Learning Algorithms in Neuroimaging: An Overview

Machine learning (ML) and artificial intelligence (AI) applications in the field of neuroimaging have been on the rise in recent years, and their clinical adoption is increasing worldwide. Deep learning (DL) is a field of ML that can be defined as a set of algorithms enabling a computer to be fed with raw data and progressively discover-through multiple layers of representation-more complex and abstract patterns in large data sets. The combination of ML and radiomics, namely the extraction of features from medical images, has proven valuable, too: Radiomic information can be used for enhanced image characterization and prognosis or outcome prediction. This chapter summarizes the basic concepts underlying ML application for neuroimaging and discusses technical aspects of the most promising algorithms, with a specific focus on Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs), in order to provide the readership with the fundamental theoretical tools to better understand ML in neuroimaging. Applications are highlighted from a practical standpoint in the last section of the chapter, including: image reconstruction and restoration, image synthesis and super-resolution, registration, segmentation, classification, and outcome prediction.

Impact of deep learning reconstruction on intracranial 1.5 T magnetic resonance angiography

Purpose

The purpose of this study was to evaluate whether deep learning reconstruction (DLR) improves the image quality of intracranial magnetic resonance angiography (MRA) at 1.5 T.

Materials and methods

In this retrospective study, MRA images of 40 patients (21 males and 19 females; mean age, 65.8 ± 13.2 years) were reconstructed with and without the DLR technique (DLR image and non-DLR image, respectively). Quantitative image analysis was performed by placing regions of interest on the basilar artery and cerebrospinal fluid in the prepontine cistern. We calculated the signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) for analyses of the basilar artery. Two experienced radiologists evaluated the depiction of structures (the right internal carotid artery, right ophthalmic artery, basilar artery, and right superior cerebellar artery), artifacts, subjective noise and overall image quality in a qualitative image analysis. Scores were compared in the quantitative and qualitative image analyses between the DLR and non-DLR images using Wilcoxon signed-rank tests.

Results

The SNR and CNR for the basilar artery were significantly higher for the DLR images than for the non-DLR images (p < 0.001). Qualitative image analysis scores (p < 0.003 and p < 0.005 for readers 1 and 2, respectively), excluding those for artifacts (p = 0.072–0.565), were also significantly higher for the DLR images than for the non-DLR images.

Conclusion

DLR enables the production of higher quality 1.5 T intracranial MRA images with improved visualization of arteries.

Introducing Swish and Parallelized Blind Removal Improves the Performance of a Convolutional Neural Network in Denoising MR Images

Purpose

To improve the performance of a denoising convolutional neural network (DnCNN) and to make it applicable to images with inhomogeneous noise, a refinement involving an activation function (AF) and an application of the refined method for inhomogeneous-noise images was examined in combination with parallelized image denoising.

Methods

Improvements in the DnCNN were performed by three approaches. One is refinement of the AF of each neural network that constructs the DnCNN. Swish was used in the DnCNN instead of rectifier linear unit. Second, blind noise removal was introduced to the DnCNN in order to adapt spatially variant noises. Third, blind noise removal was applied to parallelized image denoising, referred to herein as ParBID. The ParBID procedure is as follows: (1) adjacent 2D slice images are linearly combined to obtained higher peak SNR (PSNR) images, (2) combined images with different weight coefficients are denoised using the blind DnCNN, and (3) denoised combined images are separated into original position images by algebraic calculation.

Results

Experimental studies showed that the PSNR and the structural similarity index (SSIM) were improved by using Swish for all noise levels, from 2.5% to 7.5%, as compared to the conventional DnCNN. It was also shown that a well-trained CNN could remove spatially variant noises superimposed on images. Experimental studies with ParBID showed that the greatest PSNR and SSIM improvements were obtained at the middle slice when three slice images were used for linear image combination. More fine structures of images and image contrast remained when the proposed ParBID procedure was used.

Conclusion

Swish can improve the denoising performance of the DnCNN, and the denoising performance and effectiveness were further improved by ParBID.

Deep learning-accelerated T2-weighted imaging of the prostate: Impact of further acceleration with lower spatial resolution on image quality

PURPOSE

To compare image quality in prostate MRI among standard T2-weighted imaging (T2-std), accelerated T2-weighted imaging (T2WI) with high resolution (T2-HR) and more accelerated T2WI with lower resolution (T2-LR) using both conventional reconstruction (C) and deep learning reconstruction (DL).

MATERIALS AND METHODS

In 46 consecutive patients, T2-std, T2-HR and T2-LR were acquired in 3:32 min, 1:06 min and 0.52 min, respectively. Both reconstruction techniques (C and DL) were applied to T2-HR and T2-LR. Five sets of images (T2-std, T2-HR_C, T2-LR_C, T2-HR_DL, and T2-LR_DL) for each patient were independently evaluated by two radiologists. Quantitative analysis including the signal-to-noise ratio (SNR) and contrast ratio (CR) and qualitative analysis with a 5-point scale for the sharpness of structures, ghosting or other artifacts, noise and overall image quality were performed.

RESULTS

The SNR was not different in either the peripheral zone (PZ) or transition zone (TZ) between T2-LR_DL and T2-std with the median value of 21.7 versus 22.6 in PZ and 16.5 versus 17.3 in TZ, respectively. The CR between the prostate gland and muscle was significantly lower on T2-HR_C and T2-LR_C than on T2-std. Most of the evaluated factors showed significantly lower scores on T2-HR_C and T2-LR_C than on T2-std. Although noise and overall image quality on T2-HR_DL and other artifacts on T2-LR_DL were rated significantly lower than on T2-std (median value 4.0 versus 4.5, P < 0.001; 4.5 versus 5.0, P = 0.001; 4.5 versus 5.0, P = 0.006, respectively), other factors did not differ between T2-std and T2-HR_DL or T2-LR_DL.

CONCLUSION

DL is useful to improve image quality in accelerated T2WI of the prostate gland. Using DL, accelerated T2WI with lower spatial resolution than T2-std can be achieved with similar image quality in much shorter scan time (75.5% reduction in the acquisition time).

Breath-hold 3D magnetic resonance cholangiopancreatography at 1.5 T using a deep learning-based noise-reduction approach: Comparison with the conventional respiratory-triggered technique

OBJECTIVES

To assess the image quality of conventional respiratory-triggered 3-dimentional (3D) magnetic resonance cholangiopancreatography (Resp-MRCP) and breath-hold 3D MRCP (BH-MRCP) with and without denoising procedure using deep learning-based reconstruction (dDLR) at 1.5 T.

METHODS

Forty-two patients underwent MRCP at 1.5 T MRI. The following imaging sequences were performed: Resp-MRCP and BH-MRCP. We applied the dDLR method to the BH-MRCP data (BH-dDLR-MRCP). As a qualitative analysis, two radiologists rated the visibility of the proximal common bile duct (CBD), pancreaticobiliary junction, distal main pancreatic duct, cystic duct, and right and left hepatic ducts. Artifacts and overall image quality were also rated. The signal-to-noise ratios (SNRs), contrast ratios, and contrast-to-noise ratios (CNRs) of the CBD images were calculated for quantitative analysis.

RESULTS

BH-MRCP was successfully performed in a single BH. The qualitative and quantitative measurements for BH-dDLR-MRCP were significantly higher than for BH-MRCP (P < 0.02 and P < 0.001, respectively), and the qualitative measurements for BH-dDLR-MRCP were equivalent to or higher than for Resp-MRCP (P = 0.048-1.000). The SNRs and CNRs for BH-dDLR-MRCP were significantly higher than for Resp-MRCP (P < 0.001 and P = 0.001, respectively).

CONCLUSION

dDLR is useful and clinically feasible for BH-MRCP at 1.5 T MRI, and enables rapid imaging without loss of image quality compared to conventional Resp-MRCP.

Reconstruction of cardiovascular black-blood T2-weighted image by deep learning algorithm: A comparison with intensity filter

Background

Deep learning-based methods have been used to denoise magnetic resonance imaging.

Purpose

The purpose of this study was to evaluate a deep learning reconstruction (DL Recon) in cardiovascular black-blood T2-weighted images and compare with intensity filtered images.

Material and Methods

Forty-five DL Recon images were compared with intensity filtered and the original images. For quantitative image analysis, the signal to noise ratio (SNR) of the septum, contrast ratio (CR) of the septum to lumen, and sharpness of the endocardial border were calculated in each image. For qualitative image quality assessment, a 4-point subjective scale was assigned to each image (1 = poor, 2 = fair, 3 = good, 4 = excellent).

Results

The SNR and CR were significantly higher in the DL Recon images than in the intensity filtered and the original images (p < .05 in each). Sharpness of the endocardial border was significantly higher in the DL Recon and intensity filtered images than in the original images (p < .05 in each). The image quality of the DL Recon images was significantly better than that of intensity filtered and original images (p < .001 in each).

Conclusions

DL Recon reduced image noise while improving image contrast and sharpness in the cardiovascular black-blood T2-weight sequence.

Deep Learning and Its Application to Function Approximation for MR in Medicine: An Overview

This article presents an overview of deep learning (DL) and its applications to function approximation for MR in medicine. The aim of this article is to help readers develop various applications of DL. DL has made a large impact on the literature of many medical sciences, including MR. However, its technical details are not easily understandable for non-experts of machine learning (ML).The first part of this article presents an overview of DL and its related technologies, such as artificial intelligence (AI) and ML. AI is explained as a function that can receive many inputs and produce many outputs. ML is a process of fitting the function to training data. DL is a kind of ML, which uses a composite of many functions to approximate the function of interest. This composite function is called a deep neural network (DNN), and the functions composited into a DNN are called layers. This first part also covers the underlying technologies required for DL, such as loss functions, optimization, initialization, linear layers, non-linearities, normalization, recurrent neural networks, regularization, data augmentation, residual connections, autoencoders, generative adversarial networks, model and data sizes, and complex-valued neural networks.The second part of this article presents an overview of the applications of DL in MR and explains how functions represented as DNNs are applied to various applications, such as RF pulse, pulse sequence, reconstruction, motion correction, spectroscopy, parameter mapping, image synthesis, and segmentation.

MR Denoising Increases Radiomic Biomarker Precision and Reproducibility in Oncologic Imaging

Several noise sources, such as the Johnson-Nyquist noise, affect MR images disturbing the visualization of structures and affecting the subsequent extraction of radiomic data. We evaluate the performance of 5 denoising filters (anisotropic diffusion filter (ADF), curvature flow filter (CFF), Gaussian filter (GF), non-local means filter (NLMF), and unbiased non-local means (UNLMF)), with 33 different settings, in T2-weighted MR images of phantoms (N = 112) and neuroblastoma patients (N = 25). Filters were discarded until the most optimal solutions were obtained according to 3 image quality metrics: peak signal-to-noise ratio (PSNR), edge-strength similarity-based image quality metric (ESSIM), and noise (standard deviation of the signal intensity of a region in the background area). The selected filters were ADFs and UNLMs. From them, 107 radiomics features preservation at 4 progressively added noise levels were studied. The ADF with a conductance of 1 and 2 iterations standardized the radiomic features, improving reproducibility and quality metrics.

A Survey of Brain Tumor Segmentation and Classification Algorithms

A brain Magnetic resonance imaging (MRI) scan of a single individual consists of several slices across the 3D anatomical view. Therefore, manual segmentation of brain tumors from magnetic resonance (MR) images is a challenging and time-consuming task. In addition, an automated brain tumor classification from an MRI scan is non-invasive so that it avoids biopsy and make the diagnosis process safer. Since the beginning of this millennia and late nineties, the effort of the research community to come-up with automatic brain tumor segmentation and classification method has been tremendous. As a result, there are ample literature on the area focusing on segmentation using region growing, traditional machine learning and deep learning methods. Similarly, a number of tasks have been performed in the area of brain tumor classification into their respective histological type, and an impressive performance results have been obtained. Considering state of-the-art methods and their performance, the purpose of this paper is to provide a comprehensive survey of three, recently proposed, major brain tumor segmentation and classification model techniques, namely, region growing, shallow machine learning and deep learning. The established works included in this survey also covers technical aspects such as the strengths and weaknesses of different approaches, pre- and post-processing techniques, feature extraction, datasets, and models' performance evaluation metrics.

Quiet Diffusion-weighted MR Imaging of the Brain for Pediatric Patients with Moyamoya Disease

PURPOSE

Diffusion-weighted MRI (DWI) is an essential sequence for evaluating pediatric patients with moyamoya disease (MMD); however, acoustic noise associated with DWI may lead to motion artifact. Compared with conventional DWI (cDWI), quiet DWI (qDWI) is considered less noisy and able to keep children more relaxed and stable. This study aimed to evaluate the suitability of qDWI compared with cDWI for pediatric patients with MMD.

METHODS

In this observational study, MR examinations of the brain were performed either with or without sedation in pediatric patients with MMD between September 2017 and August 2018. Three neuroradiologists independently evaluated the images for artifacts and restricted diffusion in the brain. The differences between qDWI and cDWI were compared statistically using a chi-square test.

RESULTS

One-hundred and six MR scans of 56 patients with MMD (38 scans of 15 sedated patients: 6 boys and 9 girls; mean age, 5.2 years; range, 1-9 years; and 68 scans of 42 unsedated patients: 19 boys and 23 girls; mean age, 10.7 years; range, 7-16 years) were evaluated. MR examinations were performed either with or without sedation (except in one patient). In sedated patients, no artifact other than susceptibility was observed on qDWI, whereas four artifacts were observed on cDWI (P = .04). One patient awoke from sedation during cDWI scanning, while no patient awoke from sedation during qDWI acquisition. For unsedated patients, three scans showed artifacts on qDWI, whereas two scans showed artifacts on cDWI (P = .65). Regarding restricted diffusion, qDWI revealed three cases, while two cases were found on cDWI (P = .66).

CONCLUSION

qDWI induced fewer artifacts compared with cDWI in sedated patients, and similar frequencies of artifacts were induced by qDWI and by cDWI in unsedated patients. qDWI showed restricted diffusion comparable to cDWI.

Novel deep learning-based noise reduction technique for prostate magnetic resonance imaging

Introduction

Magnetic resonance imaging (MRI) has played an increasingly major role in the evaluation of patients with prostate cancer, although prostate MRI presents several technical challenges. Newer techniques, such as deep learning (DL), have been applied to medical imaging, leading to improvements in image quality. Our goal is to evaluate the performance of a new deep learning-based reconstruction method, “DLR” in improving image quality and mitigating artifacts, which is now commercially available as AIR^TM Recon DL (GE Healthcare, Waukesha, WI). We hypothesize that applying DLR to the T2WI images of the prostate provides improved image quality and reduced artifacts.

Methods

This study included 31 patients with a history of prostate cancer that had a multiparametric MRI of the prostate with an endorectal coil (ERC) at 1.5 T or 3.0 T. Four series of T2-weighted images were generated in total: one set with the ERC signal turned on (ERC) and another set with the ERC signal turned off (Non-ERC). Each of these sets then reconstructed using two different reconstruction methods: conventional reconstruction (Conv) and DL Recon (DLR): ERC_DLR, ERC_Conv, Non-ERC_DLR, and Non-ERC_Conv. Three radiologists independently reviewed and scored the four sets of images for (i) image quality, (ii) artifacts, and (iii) visualization of anatomical landmarks and tumor.

Results

The Non-ERC_DLR scored as the best series for (i) overall image quality (p < 0.001), (ii) reduced artifacts (p < 0.001), and (iii) visualization of anatomical landmarks and tumor.

Conclusion

Prostate imaging without the use of an endorectal coil could benefit from deep learning reconstruction as demonstrated with T2-weighted imaging MRI evaluations of the prostate.

Applicability of deep learning-based reconstruction trained by brain and knee 3T MRI to lumbar 1.5T MRI

Background

Several deep learning-based methods have been proposed for addressing the long scanning time of magnetic resonance imaging. Most are trained using brain 3T magnetic resonance images, but is unclear whether performance is affected when applying these methods to different anatomical sites and at different field strengths.

Purpose

To validate the denoising performance of deep learning-based reconstruction method trained by brain and knee 3T magnetic resonance images when applied to lumbar 1.5T magnetic resonance images.

Material and Methods

Using a 1.5T scanner, we obtained lumber T2-weighted sequences in 10 volunteers using three different scanning times: 228 s (standard), 119 s (double-fast), and 68 s (triple-fast). We compared the images obtained by the standard sequence with those obtained by the deep learning-based reconstruction-applied faster sequences.

Results

Signal-to-noise ratio values were significantly higher for deep learning-based reconstruction-double-fast than for standard and did not differ significantly between deep learning-based reconstruction-triple-fast and standard. Contrast-to-noise ratio values also did not differ significantly between deep learning-based reconstruction-triple-fast and standard. Qualitative scores for perceived signal-to-noise ratio and overall image quality were significantly higher for deep learning-based reconstruction-double fast and deep learning-based reconstruction-triple-fast than for standard. Average scores for sharpness, contrast, and structure visibility were equal to or higher for deep learning-based reconstruction-double-fast and deep learning-based reconstruction-triple-fast than for standard, but the differences were not statistically significant. The average scores for artifact were lower for deep learning-based reconstruction-double-fast and deep learning-based reconstruction-triple-fast than for standard, but the differences were not statistically significant.

Conclusion

The deep learning-based reconstruction method trained by 3T brain and knee images may reduce the scanning time of 1.5T lumbar magnetic resonance images by one-third without sacrificing image quality.

Ultrashort echo time time-spatial labeling inversion pulse magnetic resonance angiography with denoising deep learning reconstruction for the assessment of abdominal visceral arteries

Current contrast-enhanced magnetic resonance angiography (MRA) and non-contrast-enhanced balanced steady-state free precession (bSSFP) MRA cause susceptibility artifacts from metallic devices in assessing endovascular visceral-artery interventions. The aims of this study are to investigate and compare image quality (IQ) and susceptibility artifacts of three-dimensional (3D) ultrashort echo time (UTE) time-spatial labeling inversion pulse (Time-SLIP) with those of 3D bSSFP Time-SLIP and to assess denoising deep learning reconstruction (dDLR) for the improvement of the signal-to-noise ratio (SNR) in 3D UTE with sparse sampling in phantoms and human subjects. This is a prospective type of study. Pulsatile glycerin-water flow phantom with platinum-tungsten-alloy coil, stainless-steel, nitinol, and cobalt-alloy stents were used. Ten healthy volunteers (seven males) and three patients (two males) were included in this study. 3D UTE Time-SLIP and 3D bSSFP Time-SLIP at 3T were used. The phantom-based study compared the signal-intensity ratio of the device levels (SRdevice ) and distal segments (SRdistal ) to the proximal segments. The volunteer-based study measured SNR, contrast ratio (CR), and IQ. The patient study evaluated local artifacts from metallic devices. Statistical tests included paired t-tests, Wilcoxon-signed rank tests, and Kruskal-Wallis tests. In the phantom-based study, SRdevice was small with UTE Time-SLIP, except the stainless-steel stent. SRdistal was greater (49.1%-90.4%) on bSSFP images than UTE images (-11.1% to 9.6%). Among volunteers, dDLR in UTE images improved SNR (p < 0.05) and IQ (p < 0.05), but CR was unaffected. UTE Time-SLIP showed inferior SNR and IQ than bSSFP Time-SLIP in images with and without dDLR (p < 0.05 for each). However, among patients, UTE Time-SLIP showed reduced metal artifacts compared to bSSFP Time-SLIP. Irrespective of the lower SNR and IQ of 3D UTE Time-SLIP than those of 3D bSSFP Time-SLIP, the former appeared to better depict flow after stenting or coiling. This indicates the potential of 3D UTE Time-SLIP to provide suitable diagnostic images of target vessels. dDLR improved SNR with reducing artifacts related to radial sampling, while maintaining the contrast. LEVEL OF EVIDENCE: 2. TECHNICAL EFFICACY STAGE: 2.

Improvement of late gadolinium enhancement image quality using a deep learning-based reconstruction algorithm and its influence on myocardial scar quantification

OBJECTIVES

The aim of this study was to assess the effect of a deep learning (DL)-based reconstruction algorithm on late gadolinium enhancement (LGE) image quality and to evaluate its influence on scar quantification.

METHODS

Sixty patients (46 ± 17 years, 50% male) with suspected or known cardiomyopathy underwent CMR. Short-axis LGE images were reconstructed using the conventional reconstruction and a DL network (DLRecon) with tunable noise reduction (NR) levels from 0 to 100%. Image quality of standard LGE images and DLRecon images with 75% NR was scored using a 5-point scale (poor to excellent). In 30 patients with LGE, scar size was quantified using thresholding techniques with different standard deviations (SD) above remote myocardium, and using full width at half maximum (FWHM) technique in images with varying NR levels.

RESULTS

DLRecon images were of higher quality than standard LGE images (subjective quality score 3.3 ± 0.5 vs. 3.6 ± 0.7, p < 0.001). Scar size increased with increasing NR levels using the SD methods. With 100% NR level, scar size increased 36%, 87%, and 138% using 2SD, 4SD, and 6SD quantification method, respectively, compared to standard LGE images (all p values < 0.001). However, with the FWHM method, no differences in scar size were found (p = 0.06).

CONCLUSIONS

LGE image quality improved significantly using a DL-based reconstruction algorithm. However, this algorithm has an important impact on scar quantification depending on which quantification technique is used. The FWHM method is preferred because of its independency of NR. Clinicians should be aware of this impact on scar quantification, as DL-based reconstruction algorithms are being used.

KEY POINTS

• The image quality based on (subjective) visual assessment and image sharpness of late gadolinium enhancement images improved significantly using a deep learning-based reconstruction algorithm that aims to reconstruct high signal-to-noise images using a denoising technique. • Special care should be taken when scar size is quantified using thresholding techniques with different standard deviations above remote myocardium because of the large impact of these advanced image enhancement algorithms. • The full width at half maximum method is recommended to quantify scar size when deep learning algorithms based on noise reduction are used, as this method is the least sensitive to the level of noise and showed the best agreement with visual late gadolinium enhancement assessment.

Recent Advances in Parameter Inference for Diffusion MRI Signal Models

In this paper, fundamentals and recent progress for obtaining biological features quantitatively by using diffusion MRI are reviewed. First, a brief description of diffusion MRI history, application, and development was presented. Then, well-known parametric models including diffusion tensor imaging (DTI), diffusional kurtosis imaging (DKI), and neurite orientation dispersion diffusion imaging (NODDI) are introduced with several classifications in various viewpoints with other modeling schemes. In addition, this review covers mathematical generalization and examples of methodologies for the model parameter inference from conventional fitting to recent machine learning approaches, which is called Q-space learning (QSL). Finally, future perspectives on diffusion MRI parameter inference are discussed with the aspects of imaging modeling and simulation.

Feasibility of high-resolution magnetic resonance imaging of the liver using deep learning reconstruction based on the deep learning denoising technique

PURPOSE

To evaluate the feasibility of High-resolution (HR) magnetic resonance imaging (MRI) of the liver using deep learning reconstruction (DLR) based on a deep learning denoising technique compared with standard-resolution (SR) imaging.

MATERIALS AND METHODS

This retrospective study included patients who underwent abdominal MRI including both HR imaging using DLR and SR imaging between April 1 and August 31, 2019. DLR was applied to all HR images using 12 different strength levels of noise reduction to determine the optimal denoised level for HR images. The mean signal-to-noise ratio (SNR) was then compared between the original HR images without DLR and the optimal denoised HR images with DLR and SR images. The mean image noise, sharpness and overall image quality were also compared. Statistical analyses were performed with the Friedman and Dunn-Bonferroni post-hoc test.

RESULTS

In total, 49 patients were analyzed (median age, 71 years; 25 women). In quantitative analysis, the mean SNRs on the original HR images without DLR were significantly lower than those on the SR images in all sequences (p < 0.01). Conversely, the mean SNRs on optimal denoised HR images were significantly higher than those on the SR images in all sequences (p < 0.01). In the qualitative analysis, the mean scores for the image noise and overall image quality were significantly higher on optimal denoised HR images than on the SR images in all sequences (p < 0.01) except for the mean image noise score in in-phase (IP) images.

CONCLUSIONS

The use of a deep learning-based noise reduction technique substantially and successfully improved the SNR and image quality in HR imaging of the liver. Denoised HR imaging using the DLR technique appears feasible for use in liver MR examinations compared with SR imaging.

MR Imaging in the 21st Century: Technical Innovation over the First Two Decades

Clinical MRI systems have continually improved over the years since their introduction in the 1980s. In MRI technical development, the developments in each MRI system component, including data acquisition, image reconstruction, and hardware systems, have impacted the others. Progress in each component has induced new technology development opportunities in other components. New technologies outside of the MRI field, for example, computer science, data processing, and semiconductors, have been immediately incorporated into MRI development, which resulted in innovative applications. With high performance computing and MR technology innovations, MRI can now provide large volumes of functional and anatomical image datasets, which are important tools in various research fields. MRI systems are now combined with other modalities, such as positron emission tomography (PET) or therapeutic devices. These hybrid systems provide additional capabilities.

In this review, MRI advances in the last two decades will be considered. We will discuss the progress of MRI systems, the enabling technology, established applications, current trends, and the future outlook.

Deep learning-based image quality improvement of 18F-fluorodeoxyglucose positron emission tomography: a retrospective observational study

Background

Deep learning (DL)-based image quality improvement is a novel technique based on convolutional neural networks. The aim of this study was to compare the clinical value of ¹⁸F-fluorodeoxyglucose positron emission tomography (¹⁸F-FDG PET) images obtained with the DL method with those obtained using a Gaussian filter.

Methods

Fifty patients with a mean age of 64.4 (range, 19–88) years who underwent ¹⁸F-FDG PET/CT between April 2019 and May 2019 were included in the study. PET images were obtained with the DL method in addition to conventional images reconstructed with three-dimensional time of flight-ordered subset expectation maximization and filtered with a Gaussian filter as a baseline for comparison. The reconstructed images were reviewed by two nuclear medicine physicians and scored from 1 (poor) to 5 (excellent) for tumor delineation, overall image quality, and image noise. For the semi-quantitative analysis, standardized uptake values in tumors and healthy tissues were compared between images obtained using the DL method and those obtained with a Gaussian filter.

Results

Images acquired using the DL method scored significantly higher for tumor delineation, overall image quality, and image noise compared to baseline (P < 0.001). The Fleiss’ kappa value for overall inter-reader agreement was 0.78. The standardized uptake values in tumor obtained by DL were significantly higher than those acquired using a Gaussian filter (P < 0.001).

Conclusions

Deep learning method improves the quality of PET images.

Aliasing layers for processing parallel imaging and EPI ghost artifacts efficiently in convolutional neural networks

PURPOSE

The purposes of this work are to develop a method for efficiently processing MR-specific artifacts using a convolutional neural network (CNN), and to present its applications for the removal of the artifacts without suppressing actual signals. In MR images that are acquired using parallel imaging and/or EPI, the locations of aliasing artifacts and/or N-half ghost artifacts can be analytically calculated. However, existing methods using CNNs do not take the structures of the artifacts into account, and therefore need a large number of convolution layers for processing the artifacts.

METHODS

For processing the artifacts, a new layer that is named the aliasing layer (AL) is proposed. Because a CNN stands on the assumption that an image has spatial locality, a convolution layer is formulated as a linear function of neighbor locations. For processing the artifacts, the AL preprocesses MR images by moving the calculated locations to the locations accessible through summations over all channels in a standard convolution layer. To evaluate the application of ALs for the removal of parallel imaging and EPI artifacts, CNNs with ALs were compared with those without ALs.

RESULTS

The results showed that image-quality metrics of a six-layer CNN with ALs were better than those of a 12-layer CNN without ALs. The results also showed that CNNs with ALs suppressed the artifacts selectively.

CONCLUSION

The aliasing layer is proposed for processing MR-specific artifacts efficiently. The experimental results demonstrated that the AL improved CNNs for removing artifacts from parallel imaging and EPI.

Improved cortical surface reconstruction using sub-millimeter resolution MPRAGE by image denoising

Automatic cerebral cortical surface reconstruction is a useful tool for cortical anatomy quantification, analysis and visualization. Recently, the Human Connectome Project and several studies have shown the advantages of using T₁-weighted magnetic resonance (MR) images with sub-millimeter isotropic spatial resolution instead of the standard 1-mm isotropic resolution for improved accuracy of cortical surface positioning and thickness estimation. Nonetheless, sub-millimeter resolution images are noisy by nature and require averaging multiple repetitions to increase the signal-to-noise ratio for precisely delineating the cortical boundary. The prolonged acquisition time and potential motion artifacts pose significant barriers to the wide adoption of cortical surface reconstruction at sub-millimeter resolution for a broad range of neuroscientific and clinical applications. We address this challenge by evaluating the cortical surface reconstruction resulting from denoised single-repetition sub-millimeter T₁-weighted images. We systematically characterized the effects of image denoising on empirical data acquired at 0.6 mm isotropic resolution using three classical denoising methods, including denoising convolutional neural network (DnCNN), block-matching and 4-dimensional filtering (BM4D) and adaptive optimized non-local means (AONLM). The denoised single-repetition images were found to be highly similar to 6-repetition averaged images, with a low whole-brain averaged mean absolute difference of ~0.016, high whole-brain averaged peak signal-to-noise ratio of ~33.5 dB and structural similarity index of ~0.92, and minimal gray matter–white matter contrast loss (2% to 9%). The whole-brain mean absolute discrepancies in gray matter–white matter surface placement, gray matter–cerebrospinal fluid surface placement and cortical thickness estimation were lower than 165 μm, 155 μm and 145 μm—sufficiently accurate for most applications. These discrepancies were approximately one third to half of those from 1-mm isotropic resolution data. The denoising performance was equivalent to averaging ~2.5 repetitions of the data in terms of image similarity, and 1.6–2.2 repetitions in terms of the cortical surface placement accuracy. The scan-rescan variability of the cortical surface positioning and thickness estimation was lower than 170 μm. Our unique dataset and systematic characterization support the use of denoising methods for improved cortical surface reconstruction at sub-millimeter resolution.

[The Application Value of Artificial Intelligence-based Filtering and Interpolated Image Reconstruction Algorithm in Abdominal Magnetic Resonance Image Denoising]

OBJECTIVE

To compare the noise reduction performance of conventional filtering and artificial intelligence-based filtering and interpolation (AIFI) and to explore for optimal parameters of applying AIFI in the noise reduction of abdominal magnetic resonance imaging (MRI).

METHODS

Sixty patients who underwent upper abdominal MRI examination in our hospital were retrospectively included. The raw data of T1-weighted image (T1WI), T2-weighted image (T2WI), and dualecho sequences were reconstructed with two image denoising techniques, conventional filtering and AIFI of different levels of intensity. The difference in objective image quality indicators, peak signal-to-noise ratio (pSNR) and image sharpness, of the different denoising techniques was compared. Two radiologists evaluated the image noise, contrast, sharpness, and overall image quality. Their scores were compared and the interobserver agreement was calculated.

RESULTS

Compared with the original images, improvement of varying degrees were shown in the pSNR and the sharpness of the images of the three sequences, T1W1, T2W2, and dual echo sequence, after denoising filtering and AIFI were used (all P<0.05). In addition, compared with conventional filtering, the objective quality scores of the reconstructed images were improved when conventional filtering was combined with AIFI reconstruction methods in T1WI sequence, AIFI level≥3 was used in T2WI and echo1 sequence, and AIFI level≥4 was used in echo2 sequence (all P<0.05). The subjective scores given by the two radiologists for the image noise, contrast, sharpness, and overall image quality in each sequence of conventional filtering reconstruction, AIFI reconstruction (except for AIFI level=1), and two-method combination reconstruction were higher than those of the original images (all P<0.05). However, the image contrast scores were reduced for AIFI level=5. There was good interobserver agreement between the two radiologists (all r>0.75, P<0.05). After multidimensional comparison, the optimal parameters of using AIFI technique for noise reduction in abdominal MRI were conventional filtering+AIFI level=3 in the T1WI sequence and AIFI level=4 in the T2WI and dualecho sequences.

CONCLUSION

AIFI is superior to filtering in imaging denoising at medium and high levels. It is a promising noise reduction technique. The optimal parameters of using AIFI for abdominal MRI are Filtering+AIFI level=3 in the T1WI sequence and AIFI level=4 in T2WI and dualecho sequences.

Compressed sensing and deep learning reconstruction for women's pelvic MRI denoising: Utility for improving image quality and examination time in routine clinical practice

PURPOSE

To demonstrate the utility of compressed sensing with parallel imaging (Compressed SPEEDER) and AiCE compared with that of conventional parallel imaging (SPEEDER) for shortening examination time and improving image quality of women's pelvic MRI.

METHOD

Thirty consecutive patients with women's pelvic diseases (mean age 50 years) underwent T2-weighted imaging using Compressed SPEEDER as well as conventional SPEEDER reconstructed with and without AiCE. The examination times were recorded, and signal-to-noise ratio (SNR) was calculated for every patient. Moreover, overall image quality was assessed using a 5-point scoring system, and final scores for all patients were determined by consensus of two readers. Mean examination time, SNR and overall image quality were compared among the four data sets by Wilcoxon signed-rank test.

RESULTS

Examination times for Compressed SPEEDER with and without AiCE were significantly shorter than those for conventional SPEEDER with and without AiCE (with AiCE: p < 0.0001, without AiCE: p < 0.0001). SNR of Compressed SPEEDER and of SPEEDER with AiCE was significantly superior to that of Compressed SPEEDER without AiCE (vs. Compressed SPEEDER, p = 0.01; vs. SPEEDER, p = 0.009). Overall image quality of Compressed SPEEDER with AiCE and of SPEEDER with and without AiCE was significantly higher than that of Compressed SPEEDER without AiCE (vs. Compressed SPEEDER with AiCE, p < 0.0001; vs. SPEEDER with AiCE, p < 0.0001; SPEEDER without AiCE, p = 0.0003).

CONCLUSION

Image quality and shorten examination time for T2-weighted imaging in women's pelvic MRI can be significantly improved by using Compressed SPEEDER with AiCE in comparison with conventional SPEEDER, although other sequences were not tested.

Image quality improvement of single-shot turbo spin-echo magnetic resonance imaging of female pelvis using a convolutional neural network

We have developed a deep learning-based approach to improve image quality of single-shot turbo spin-echo (SSTSE) images of female pelvis. We aimed to compare the deep learning-based single-shot turbo spin-echo (DL-SSTSE) images of female pelvis with turbo spin-echo (TSE) and conventional SSTSE images in terms of image quality.One hundred five and 21 subjects were used as training and test sets, respectively. We performed 6-fold cross validation. In the training process, low-quality images were generated from TSE images as input. TSE images were used as ground truth images. In the test process, the trained convolutional neural network was applied to SSTSE images. The output images were denoted as DL-SSTSE images. Apart from DL-SSTSE images, classical filtering methods were adopted to SSTSE images. Generated images were denoted as F-SSTSE images. Contrast ratio (CR) of gluteal fat and myometrium and signal-to-noise ratio (SNR) of gluteal fat were measured for all images. Two radiologists graded these images using a 5-point scale and evaluated the image quality with regard to overall image quality, contrast, noise, motion artifact, boundary sharpness of layers in the uterus, and the conspicuity of the ovaries. CRs, SNRs, and image quality scores were compared using the Steel-Dwass multiple comparison tests.CRs and SNRs were significantly higher in DL-SSTSE, F-SSTSE, and TSE images than in SSTSE images. Scores with regard to overall image quality, contrast, noise, and boundary sharpness of layers in the uterus were significantly higher on DL-SSTSE and TSE images than on SSTSE images. There were no significant differences in the CRs, SNRs, and respective scores between DL-SSTSE and TSE images. The score with regard to motion artifacts was significantly higher on DL-SSTSE, F-SSTSE, and SSTSE images than on TSE images. The score with regard to the conspicuity of ovaries was significantly higher on DL-SSTSE images than on F-SSTSE, SSTSE, and TSE images (P < .001).DL-SSTSE images showed higher image quality as compared with SSTSE images. In comparison with conventional TSE images, DL-SSTSE images had acceptable image quality while keeping the advantage of the motion artifact-robustness and acquisition time efficiency in SSTSE imaging.

Higher b-values improve the correlation between diffusion MRI and the cortical microarchitecture

PURPOSE

In diffusion MRI (dMRI), it remains unclear to know how much increase of b-value is conveying additional biological meaning. We tested the correlations between cortical microarchitecture and diffusion metrics computed from standard (1000 s/mm²), high (3000 s/mm²), to very high (5000 s/mm²) b-value dMRI.

METHODS

Healthy volunteers were scanned with a dMRI pulse sequence that was first optimized together with a T1-WI and T2-WI. Averaged cortical surface map of estimated myelin (T1-WI/T2-WI) was compared with surface maps of mean diffusivity (MD) computed from each b-value (MD₁₀₀₀, MD₃₀₀₀, and MD₅₀₀₀) and to surface map of mean kurtosis (MK computed from the 0-, 1000-, to 3000-s/mm² shells) in 360 cortical parcels using Spearman correlations, multiple linear regressions, and Akaike information criteria (AIC).

RESULTS

Surface map from MD₁₀₀₀ showed variations not related to myelin but the MD₃₀₀₀ and MD₅₀₀₀ maps inversely mirrored estimated myelin map; lower MD values being observed in more myelinated cortical areas. MK mirrored myelinated cortical areas. Quantitatively, Spearman correlations between myelin and MD became more and more negative as long as b-values increased while the correlation was positive between myelin and MK. Multiple regression models confirmed negative associations between myelin and MD that were significantly better from MD₁₀₀₀ to MD₃₀₀₀ and MD₅₀₀₀ (R² = 0.33, p < 0.001; R² = 0.43, p < 0.001; and R² = 0.50, p < 0.001) and positive association between myelin and MK (R² = 0.53, p < 0.001). Comparisons of the 3 statistical models showed the best performances with MK and MD₅₀₀₀ (AIC_MK < AIC_MD5000 < AIC_MD3000 < AIC_MD1000).

CONCLUSION

Higher b-values are more closely related to subtle cellular variations of the cortical microarchitecture.

Improved Diagnostic Accuracy of Bone Metastasis Detection by Water-HAP Associated to Non-Contrast CT

We examined whether water-hydroxyapatite (HAP) images improve the diagnostic accuracy of bone metastasis compared with non-contrast CT alone. We retrospectively evaluated dual-energy computed tomography (DECT) images of 83 cancer patients (bone metastasis, 31; without bone metastasis, 52) from May 2018 to June 2019. Initially, two evaluators examined for bone metastasis on conventional CT images. In the second session, both CT and CT images plus water-HAP images on DECT. The confidence of bone metastasis was scored from 1 (benign) to 5 (malignant). The sensitivity, specificity, positive predictive values, and negative predictive values for both modalities were calculated based on true positive and negative findings. The intra-observer area under curve (AUC) for detecting bone metastasis was compared by receiver operating characteristic analysis. Kappa coefficient calculated the inter-observer agreement. In conventional CT images, sensitivity, specificity, positive predictive value, and negative predictive value of raters 1 and 2 for the identification of bone metastases were 0.742 and 0.710, 0.981 and 0.981, 0.958 and 0.957, and 0.864 and 0.850, respectively. In water-HAP, they were 1.00 and 1.00, 0.981 and 1.00, 0.969 and 1.00, and 1.00 and 1.00, respectively. In CT, AUCs were 0.861 and 0.845 in each observer. On water-HAP images, AUCs were 0.990 and 1.00. Kappa coefficient was 0.964 for CT and 0.976 for water-HAP images. The combination of CT and water-HAP images significantly increased diagnostic accuracy for detecting bone metastasis. Water-HAP images on DECT may enable accurate initial staging, reduced radiation exposure, and cost.

Non-contrast coronary magnetic resonance angiography: current frontiers and future horizons

Coronary magnetic resonance angiography (coronary MRA) is advantageous in its ability to assess coronary artery morphology and function without ionizing radiation or contrast media. However, technical limitations including reduced spatial resolution, long acquisition times, and low signal-to-noise ratios prevent it from clinical routine utilization. Nonetheless, each of these limitations can be specifically addressed by a combination of novel technologies including super-resolution imaging, compressed sensing, and deep-learning reconstruction. In this paper, we first review the current clinical use and motivations for non-contrast coronary MRA, discuss currently available coronary MRA techniques, and highlight current technical developments that hold unique potential to optimize coronary MRA image acquisition and post-processing. In the final section, we examine the various research-based coronary MRA methods and metrics that can be leveraged to assess coronary stenosis severity, physiological function, and atherosclerotic plaque characterization. We specifically discuss how such technologies may contribute to the clinical translation of coronary MRA into a robust modality for routine clinical use.

Deep Learning-based Noise Reduction for Fast Volume Diffusion Tensor Imaging: Assessing the Noise Reduction Effect and Reliability of Diffusion Metrics

To assess the feasibility of a denoising approach with deep learning-based reconstruction (dDLR) for fast volume simultaneous multi-slice diffusion tensor imaging of the brain, noise reduction effects and the reliability of diffusion metrics were evaluated with 20 patients. Image noise was significantly decreased with dDLR. Although fractional anisotropy (FA) of deep gray matter was overestimated when the number of image acquisitions was one (NAQ1), FA in NAQ1 with dDLR became closer to that in NAQ5.

Deep learning of brain magnetic resonance images: A brief review

Magnetic resonance imaging (MRI) is one of the most popular techniques in brain science and is important for understanding brain function and neuropsychiatric disorders. However, the processing and analysis of MRI is not a trivial task with lots of challenges. Recently, deep learning has shown superior performance over traditional machine learning approaches in image analysis. In this survey, we give a brief review of the recent popular deep learning approaches and their applications in brain MRI analysis. Furthermore, popular brain MRI databases and deep learning tools are also introduced. The strength and weaknesses of different approaches are addressed, and challenges as well as future directions are also discussed.

MR Imaging of Endolymphatic Hydrops: Utility of iHYDROPS-Mi2 Combined with Deep Learning Reconstruction Denoising

Purpose:

MRI of endolymphatic hydrops (EH) 4 h after intravenous administration of a single dose of gadolinium-based contrast agent is used for clinical examination in some institutions; however, further improvement in image quality would be valuable for wider clinical utility. Denoising using deep learning reconstruction (Advanced Intelligent Clear-IQ Engine [AiCE]) has been reported for CT and MR. The purpose of this study was to compare the contrast-to-noise ratio of endolymph to perilymph (CNR_EP) between the improved hybrid of reversed image of the positive endolymph signal and the native image of the perilymph signal multiplied with the heavily T₂-weighted MR cisternography (iHYDROPS-Mi2) images, which used AiCE for the three source images (i.e. positive endolymph image [PEI], positive perilymph image [PPI], MR cisternography [MRC]) to those that did not use AiCE. We also examined if there was a difference between iHYDROPS-Mi2 images with and without AiCE for degree of visual grading of EH and in endolymphatic area [EL] ratios.

Methods:

Nine patients with suspicion of EH were imaged on a 3T MR scanner. iHYDROPS images were generated by subtraction of PEI images from PPI images. iHYDROPS-Mi2 images were then generated by multiplying MRC with iHYDROPS images. The CNR_EP and EL ratio were measured on the iHYDROPS-Mi2 images. Degree of radiologist visual grading for EH was evaluated.

Results:

Mean CNR_EP ± standard deviation was 1681.8 ± 845.2 without AiCE and 7738.6 ± 5149.2 with AiCE (P = 0.00002). There was no significant difference in EL ratio for images with and without AiCE. Radiologist grading for EH agreed completely between the 2 image types in both the cochlea and vestibule.

Conclusion:

The CNR_EP of iHYDROPS-Mi2 images with AiCE had more than a fourfold increase compared with that without AiCE. Use of AiCE did not adversely affect radiologist grading of EH.

A preliminary study of deep learning-based reconstruction specialized for denoising in high-frequency domain: usefulness in high-resolution three-dimensional magnetic resonance cisternography of the cerebellopontine angle

PURPOSE

Deep learning-based reconstruction (DLR) has been developed to reduce image noise and increase the signal-to-noise ratio (SNR). We aimed to evaluate the efficacy of DLR for high spatial resolution (HR)-MR cisternography.

METHODS

This retrospective study included 35 patients who underwent HR-MR cisternography. The images were reconstructed with or without DLR. The SNRs of the CSF and pons, contrast of the CSF and pons, and sharpness of the normal-side trigeminal nerve using full width at half maximum (FWHM) were compared between the two image types. Noise quality, sharpness, artifacts, and overall image quality of these two types of images were qualitatively scored.

RESULTS

The SNRs of the CSF and pons were significantly higher with DLR than without DLR (CSF 21.81 ± 7.60 vs. 15.33 ± 4.03, p < 0.001; pons 5.96 ± 1.38 vs. 3.99 ± 0.48, p < 0.001). There were no significant differences in the contrast of the CSF and pons (p = 0.225) and sharpness of the normal-side trigeminal nerve using FWHM (p = 0.185) without and with DLR, respectively. Noise quality and the overall image quality were significantly higher with DLR than without DLR (noise quality 3.95 ± 0.19 vs. 2.53 ± 0.44, p < 0.001; overall image quality 3.97 ± 0.17 vs. 2.97 ± 0.12, p < 0.001). There were no significant differences in sharpness (p = 0.371) and artifacts (p = 1) without and with DLR.

CONCLUSION

DLR can improve the image quality of HR-MR cisternography by reducing image noise without sacrificing contrast or sharpness.

Compressed-Sensing Magnetic Resonance Image Reconstruction Using an Iterative Convolutional Neural Network Approach

Convolutional neural networks (CNNs) demonstrate excellent performance when employed to reconstruct the images obtained by compressed-sensing magnetic resonance imaging (CS-MRI). Our study aimed to enhance image quality by developing a novel iterative reconstruction approach that utilizes image-based CNNs and k-space correction to preserve original k-space data. In the proposed method, CNNs represent a priori information concerning image spaces. First, the CNNs are trained to map zero-filling images onto corresponding full-sampled images. Then, they recover the zero-filled part of the k-space data. Subsequently, k-space corrections, which involve the replacement of unfilled regions by original k-space data, are implemented to preserve the original k-space data. The above-mentioned processes are used iteratively. The performance of the proposed method was validated using a T2-weighted brain-image dataset, and experiments were conducted with several sampling masks. Finally, the proposed method was compared with other noniterative approaches to demonstrate its effectiveness. The aliasing artifacts in the reconstructed images obtained using the proposed approach were reduced compared to those using other state-of-the-art techniques. In addition, the quantitative results obtained in the form of the peak signal-to-noise ratio and structural similarity index demonstrated the effectiveness of the proposed method. The proposed CS-MRI method enhanced MR image quality with high-throughput examinations.

Non-Local SVD Denoising of MRI Based on Sparse Representations

Magnetic Resonance (MR) Imaging is a diagnostic technique that produces noisy images, which must be filtered before processing to prevent diagnostic errors. However, filtering the noise while keeping fine details is a difficult task. This paper presents a method, based on sparse representations and singular value decomposition (SVD), for non-locally denoising MR images. The proposed method prevents blurring, artifacts, and residual noise. Our method is composed of three stages. The first stage divides the image into sub-volumes, to obtain its sparse representation, by using the KSVD algorithm. Then, the global influence of the dictionary atoms is computed to upgrade the dictionary and obtain a better reconstruction of the sub-volumes. In the second stage, based on the sparse representation, the noise-free sub-volume is estimated using a non-local approach and SVD. The noise-free voxel is reconstructed by aggregating the overlapped voxels according to the rarity of the sub-volumes it belongs, which is computed from the global influence of the atoms. The third stage repeats the process using a different sub-volume size for producing a new filtered image, which is averaged with the previously filtered images. The results provided show that our method outperforms several state-of-the-art methods in both simulated and real data.

Accelerated Acquisition of High-resolution Diffusion-weighted Imaging of the Brain with a Multi-shot Echo-planar Sequence: Deep-learning-based Denoising

To accelerate high-resolution diffusion-weighted imaging with a multi-shot echo-planar sequence, we propose an approach based on reduced averaging and deep learning. Denoising convolutional neural networks can reduce amplified noise without requiring extensive averaging, enabling shorter scan times and high image quality. The preliminary experimental results demonstrate the superior performance of the proposed denoising method over state-of-the-art methods such as the widely used block-matching and 3D filtering.

Effects of Deep Learning Reconstruction Technique in High-Resolution Non-contrast Magnetic Resonance Coronary Angiography at a 3-Tesla Machine

Purpose:

To evaluate the effects of deep learning reconstruction (DLR) in qualitative and quantitative image quality of non-contrast magnetic resonance coronary angiography (MRCA).

Methods:

Ten healthy volunteers underwent conventional MRCA (C-MRCA) and high-resolution (HR) MRCA on a 3T magnetic resonance imaging with a voxel size of 1.8 × 1.1 × 1.7 mm3 and 1.8 × 0.6 × 1.0 mm3, respectively, for C-MRCA and HR-MRCA. High-resolution magnetic resonance coronary angiography was also reconstructed with the DLR technique (DLR-HR-MRCA). We compared the contrast-to-noise ratio (CNR) and visual evaluation scores for vessel sharpness and traceability of proximal and distal coronary vessels on a 4-point scale among 3 image series.

Results:

The vascular CNR value on the C-MRCA and the DLR-HR-MRCA was significantly higher than that on the HR-MRCA in the proximal and distal coronary arteries (13.9 ± 6.4, 11.3 ± 4.4, and 7.8 ± 2.6 for C-MRCA, DLR-HR-MRCA, and HR-MRCA, P < .05, respectively). Mean visual evaluation scores for the vessel sharpness and traceability of proximal and distal coronary vessels were significantly higher on the HR-DLR-MRCA than the C-MRCA ( P < .05, respectively).

Conclusion:

Deep learning reconstruction significantly improved the CNR of coronary arteries on HR-MRCA, resulting in both higher visual image quality and better vessel traceability compared with C-MRCA.