Accelerating FLAIR imaging via deep learning reconstruction: potential for evaluating white matter hyperintensities.

PURPOSE: To evaluate deep learning-reconstructed (DLR)-fluid-attenuated inversion recovery (FLAIR) images generated from undersampled data, compare them with fully sampled and rapidly acquired FLAIR images, and assess their potential for white matter hyperintensity evaluation. MATERIALS AND METHODS: We examined 30 patients with white matter hyperintensities, obtaining fully sampled FLAIR images (standard FLAIR, std-FLAIR). We created accelerated FLAIR (acc-FLAIR) images using one-third of the fully sampled data and applied deep learning to generate DLR-FLAIR images. Three neuroradiologists assessed the quality (amount of noise and gray/white matter contrast) in all three image types. The reproducibility of hyperintensities was evaluated by comparing a subset of 100 hyperintensities in acc-FLAIR and DLR-FLAIR images with those in the std-FLAIR images. Quantitatively, similarities and errors of the entire image and the focused regions on white matter hyperintensities in acc-FLAIR and DLR-FLAIR images were measured against std-FLAIR images using structural similarity index measure (SSIM), regional SSIM, normalized root mean square error (NRMSE), and regional NRMSE values. RESULTS: All three neuroradiologists evaluated DLR-FLAIR as having significantly less noise and higher image quality scores compared with std-FLAIR and acc-FLAIR (p < 0.001). All three neuroradiologists assigned significantly higher frontal lobe gray/white matter visibility scores for DLR-FLAIR than for acc-FLAIR (p < 0.001); two neuroradiologists attributed significantly higher scores for DLR-FLAIR than for std-FLAIR (p < 0.05). Regarding white matter hyperintensities, all three neuroradiologists significantly preferred DLR-FLAIR (p < 0.0001). DLR-FLAIR exhibited higher similarity to std-FLAIR in terms of visibility of the hyperintensities, with 97% of the hyperintensities rated as nearly identical or equivalent. Quantitatively, DLR-FLAIR demonstrated significantly higher SSIM and regional SSIM values than acc-FLAIR, with significantly lower NRMSE and regional NRMSE values (p < 0.0001). CONCLUSIONS: DLR-FLAIR can reduce scan time and generate images of similar quality to std-FLAIR in patients with white matter hyperintensities. Therefore, DLR-FLAIR may serve as an effective method in traditional magnetic resonance imaging protocols.

Automated Detection of Cerebral Microbleeds on Two-dimensional Gradient-recalled Echo T2* Weighted Images Using a Morphology Filter Bank and Convolutional Neural Network.

PURPOSE: We present a novel algorithm for the automated detection of cerebral microbleeds (CMBs) on 2D gradient-recalled echo T2* weighted images (T2*WIs). This approach combines a morphology filter bank with a convolutional neural network (CNN) to improve the efficiency of CMB detection. A technical evaluation was performed to ascertain the algorithm's accuracy. METHODS: In this retrospective study, 60 patients with CMBs on T2*WIs were included. The gold standard was set by three neuroradiologists based on the Microbleed Anatomic Rating Scale guidelines. Images with CMBs were extracted from the training dataset comprising 30 cases using a morphology filter bank, and false positives (FPs) were removed based on the threshold of size and signal intensity. The extracted images were used to train the CNN (Vgg16). To determine the effectiveness of the morphology filter bank, the outcomes of the following two methods for detecting CMBs from the 30-case test dataset were compared: (a) employing the morphology filter bank and additional FP removal and (b) comprehensive detection without filters. The trained CNN processed both sets of initial CMB candidates, and the final CMB candidates were compared with the gold standard. The sensitivity and FPs per patient of both methods were compared. RESULTS: After CNN processing, the morphology-filter-bank-based method had a 95.0% sensitivity with 4.37 FPs per patient. In contrast, the comprehensive method had a 97.5% sensitivity with 25.87 FPs per patient. CONCLUSION: Through effective CMB candidate refinement with a morphology filter bank and FP removal with a CNN, we achieved a high CMB detection rate and low FP count. Combining a CNN and morphology filter bank may facilitate the accurate automated detection of CMBs on T2*WIs.

Nature of the Intracellular-contrast-enhancing Fat-saturated T1-weighted Gradient-echo (ICE-TIGRE) Sequence: A Fat-suppressed T1-weighted Technique with Motion-sensitised Driven-equilibrium for Improved Contrast Enhancement in Liver Imaging.

Gadoxetic acid is both an extracellular- and hepatocyte-specific contrast agent. Signals from the extracellular space may lower the contrast between lesions and the surrounding hepatic parenchyma. To improve hepatocyte-specific enhancement, we developed an intracellular contrast-enhancing fat-saturated T1-weighted gradient-echo nature of the sequence (ICE-TIGRE). It incorporates the motion-sensitized driven-equilibrium (MSDE) pulse to suppress signals from the blood flow. We investigated the optimal ICE-TIGRE scanning parameters, i.e., the order of the MSDE and the fat saturation pulses, the duration time, and the b value of the MSDE pulse, using a phantom and three volunteers without applying gadoxetic acid. ICE-TIGRE successfully increased the contrast between the liver parenchyma and the portal vein. To maintain fat saturation, the preparation pulse order should be MSDE-fat saturation. A duration time of 21 ms should be applied to minimize the effect of the T2 factor on the T1 contrast, and a b value of 60 s/mm2 should be applied to maximize the diffusion contrast for ICE-TIGRE with the imaging system used in this study.

Histological Properties of a Chemically Fixed Human Embryo Visualized with Quantitative Susceptibility Mapping.

A chemically fixed Carnegie stage 23 (approximately 56 days of gestation) human embryo specimen was imaged using 3D spin-echo and gradient-echo sequences in a static magnetic field strength of 4.74T, and a quantitative susceptibility map was calculated using the 3D gradient-echo image. The acquired 3D microscopic images (90 µm cube voxel size) clarified the relationship between R2 (transverse relaxation rate), R2* (apparent transverse relaxation rate), and magnetic susceptibility in the heart, liver, kidney, and spinal cord. The results suggested that the R2* and magnetic susceptibility in each tissue were probably due to paramagnetic iron ions originating from erythrocytes. The large R2* (~130 s-1) and magnetic susceptibility (~0.122 ppm) in the liver were attributed to its hemopoietic function. A large magnetic susceptibility (~0.116 ppm) was also observed in the spinal cord, but we conclude that more detailed future studies are needed.

Prediction of amyloid positron emission tomography positivity using multiple regression analysis of quantitative susceptibility mapping.

PURPOSE: To develop a method for predicting amyloid positron emission tomography (PET) positivity based on multiple regression analysis of quantitative susceptibility mapping (QSM). MATERIALS AND METHODS: This prospective study included 39 patients with suspected dementia from four centers. QSM images were obtained through a 3-T, three-dimensional radiofrequency-spoiled gradient-echo sequence with multiple echoes. The cortical standard uptake value ratio (SUVR) was obtained using amyloid PET with 18F-flutemetamol, and susceptibility in the brain regions was obtained using QSM. A multiple regression model to predict cortical SUVR was constructed based on susceptibilities in multiple brain regions, with the constraint that cortical SUVR and susceptibility were positively correlated. The discrimination performance of the Aß-positive and Aß-negative cohorts was evaluated based on the predicted SUVR using the area under the receiver operating characteristic curve (AUC) and Mann-Whitney U test. RESULTS: The correlation coefficients between true and predicted SUVR were increased by incorporating the constraint, and the AUC to discriminate between the Aß-positive and Aß-negative cohorts reached to 0.79 (p < 0.01). CONCLUSION: These preliminary results suggest that a QSM-based multiple regression model can predict amyloid PET positivity with fair accuracy.

Numerical and Clinical Evaluation of the Robustness of Open-source Networks for Parallel MR Imaging Reconstruction.

PURPOSE: Deep neural networks (DNNs) for MRI reconstruction often require large datasets for training. Still, in clinical settings, the domains of datasets are diverse, and how robust DNNs are to domain differences between training and testing datasets has been an open question. Here, we numerically and clinically evaluate the generalization of the reconstruction networks across various domains under clinically practical conditions and provide practical guidance on what points to consider when selecting models for clinical application. METHODS: We compare the reconstruction performance between four network models: U-Net, the deep cascade of convolutional neural networks (DC-CNNs), Hybrid Cascade, and variational network (VarNet). We used the public multicoil dataset fastMRI for training and testing and performed a single-domain test, where the domains of the dataset used for training and testing were the same, and cross-domain tests, where the source and target domains were different. We conducted a single-domain test (Experiment 1) and cross-domain tests (Experiments 2-4), focusing on six factors (the number of images, sampling pattern, acceleration factor, noise level, contrast, and anatomical structure) both numerically and clinically. RESULTS: U-Net had lower performance than the three model-based networks and was less robust to domain shifts between training and testing datasets. VarNet had the highest performance and robustness among the three model-based networks, followed by Hybrid Cascade and DC-CNN. Especially, VarNet showed high performance even with a limited number of training images (200 images/10 cases). U-Net was more robust to domain shifts concerning noise level than the other model-based networks. Hybrid Cascade showed slightly better performance and robustness than DC-CNN, except for robustness to noise-level domain shifts. The results of the clinical evaluations generally agreed with the results of the quantitative metrics. CONCLUSION: In this study, we numerically and clinically evaluated the robustness of the publicly available networks using the multicoil data. Therefore, this study provided practical guidance for clinical applications.

Utility of Wavelet Denoising with Geometry Factor Weighting for Gadoxetic Acid-enhanced Hepatobiliary-phase MR Imaging.

PURPOSE: The wavelet denoising with geometry factor weighting (g-denoising) method can reduce the image noise by adapting to spatially varying noise levels induced by parallel imaging. The aim of this study was to investigate the clinical applicability of g-denoising on hepatobiliary-phase (HBP) images with gadoxetic acid. METHODS: We subjected 53 patients suspected of harboring hepatic neoplastic lesions to gadoxetic acid-enhanced HBP imaging with and without g-denoising (g+HBP and g-HBP). The matrix size was reduced for g+HBP images to avoid prolonging the scanning time. Two radiologists calculated the SNR, the portal vein-, and paraspinal muscle contrast-to-noise ratio (CNR) relative to the hepatic parenchyma (liver-to-portal vein- and liver-to-muscle CNR). Two other radiologists independently graded the sharpness of the liver edge, the visibility of intrahepatic vessels, the image noise, the homogeneity of liver parenchyma, and the overall image quality using a 5-point scale. Differences between g-HBP and g+HBP images were determined with the two-sided Wilcoxon signed-rank test. RESULTS: The liver-to-portal- and liver-to-muscle CNR and the SNR were significantly higher on g+HBP- than g-HBP images (P < 0.01), as was the qualitative score for the image noise, homogeneity of liver parenchyma, and overall image quality (P < 0.01). Although there were no significant differences in the scores for the sharpness of the liver edge or the score assigned for the visibility of intrahepatic vessels (P = 0.05, 0.43), with g+HBP the score was lower in three patients for the sharpness of the liver edge and in six patients for the visibility of intrahepatic vessels. CONCLUSION: At gadoxetic acid-enhanced HBP imaging, g-denoising yielded a better image quality than conventional HBP imaging although the anatomic details may be degraded.

Region Expansion of Background Field Removal with Local Spherical Harmonics Approximation for Whole-brain Quantitative Susceptibility Mapping.

PURPOSE: Quantitative susceptibility mapping (QSM) is useful for obtaining biological information. To calculate susceptibility distribution, it is necessary to calculate the local field caused by the differences of susceptibility between the tissues. The local field can be obtained by removing a background field from a total field acquired by MR phase image. Conventional approaches based on spherical mean value (SMV) filtering, which are widely used for background field calculations, fail to calculate the background field of the brain surface region corresponding to the radius of the SMV kernel, and consequently cannot calculate the QSM of the brain surface region. Accordingly, a new method calculating the local field by expansively removing the background field is proposed for whole brain QSM. METHODS: The proposed method consists of two steps. First, the background field of the brain surface is calculated from the total field using a locally polynomial approximation of spherical harmonics. Second, the whole brain local field is calculated by SMV filtering with a constraint term of the background field of the brain surface. The parameters of the approximation were optimized to reduce calculation errors through simulations using both a numerical phantom and a measured human brain. Performance of the proposed method with the optimized parameters was quantitatively and visually compared with conventional methods in an experiment of five healthy volunteers. RESULTS: The proposed method showed the accurate local field over the expanded brain region in the simulation studies. It also showed consistent QSM with conventional methods inside of the brain surface and showed clear vein structures on the brain surface. CONCLUSION: The proposed method enables accurate calculation of whole brain QSM without eroding the brain surface region while maintaining same values inside of the brain surface as the conventional methods.

Efficacy of Quantitative Susceptibility Mapping with Brain Surface Correction and Vein Removal for Detecting Increase Magnetic Susceptibility in Patients with Alzheimer's Disease.

PURPOSE: Studies on quantitative susceptibility mapping (QSM) have reported an increase in magnetic susceptibilities in patients with Alzheimer's disease (AD). Despite the pathological importance of the brain surface areas, they are sometimes excluded in QSM analysis. This study aimed to reveal the efficacy of QSM analysis with brain surface correction (BSC) and/or vein removal (VR) procedures. METHODS: Thirty-seven AD patients and 37 age- and sex-matched, cognitively normal (CN) subjects were included. A 3D-gradient echo sequence at 3T MRI was used to obtain QSM. QSM images were created with regularization enabled sophisticated harmonic artifact reduction for phase data (RESHARP) and constrained RESHARP with BSC and/or VR. We conducted ROI analysis between AD patients and CN subjects who did or did not undergo BSC and/or VR using a t-test, to compare the susceptibility values after gray matter weighting. RESULTS: The susceptibility values in RESHARP without BSC were significantly larger in AD patients than in CN subjects in one region (precentral gyrus, 8.1 ± 2.9 vs. 6.5 ± 2.1 ppb) without VR and one region with VR (precentral gyrus, 7.5 ± 2.8 vs. 5.9 ± 2.0 ppb). Three regions in RESHARP with BSC had significantly larger susceptibilities without VR (precentral gyrus, 7.1 ± 2.0 vs. 5.9 ± 2.0 ppb; superior medial frontal gyrus, 5.7 ± 2.6 vs. 4.2 ± 3.1 ppb; putamen, 47,8 ± 16.5 vs. 40.0 ± 15.9 ppb). In contrast, six regions showed significantly larger susceptibilities with VR in AD patients than in CN subjects (precentral gyrus, 6.4 ± 1.9 vs. 4.9 ± 2.7 ppb; superior medial frontal gyrus, 5.3 ± 2.7 vs. 3.7 ± 3.3 ppb; orbitofrontal cortex, -2.1 ± 2.7 vs. -3.6 ± 3.2 ppb; parahippocampal gyrus, 0.1 ± 3.6 vs. -1.7 ± 3.7 ppb; putamen, 45.0 ± 14.9 vs. 37.6 ± 14.6 ppb; inferior temporal gyrus, -3.4 ± 1.5 vs. -4.4 ± 1.5 ppb). CONCLUSION: RESHARP with BSC and VR showed more regions of increased susceptibility in AD patients than in CN subjects. This study highlights the efficacy of this method in facilitating the diagnosis of AD.

Three-dimensional Multi-parameter Mapping of Relaxation Times and Susceptibility Using Partially RF-spoiled Gradient Echo.

PURPOSE: MR parameter mapping is a technique that obtains distributions of parameters such as relaxation time and proton density (PD) and is starting to be used for disease quantification in clinical diagnoses. Quantitative susceptibility mapping is also promising for the early diagnosis of brain disorders such as degenerative neurological disorders. Therefore, we developed an MR quantitative parameter mapping (QPM) method to map four tissue-related parameters (T1, T2*, PD, and susceptibility) and B1 simultaneously by using a 3D partially RF-spoiled gradient echo (pRSGE). We verified the accuracy and repeatability of QPM in phantom and volunteer experiments. METHODS: Tissue-related parameters are estimated by varying four scan parameters of the 3D pRSGE: flip angle, RF-pulse phase increment, TR and TE, performing multiple image scans, and finding a least-squares fit for an intensity function (which expresses the relationship between the scan parameters and intensity values). The intensity function is analytically complex, but by using a Bloch simulation to create it numerically, the least-squares fitting can be used to estimate the quantitative values. This has the advantage of shortening the image-reconstruction processing time needed to estimate the quantitative values than with methods using pattern matching. RESULTS: A 1.1-mm isotropic resolution scan covering the whole brain was completed with a scan time of approximately 12 minutes, and the reconstruction time using a GPU was approximately 1 minute. The phantom experiments confirmed that both the accuracy and repeatability of the quantitative values were high. The volunteer scans also confirmed that the accuracy of the quantitative values was comparable to that of conventional methods. CONCLUSION: The proposed QPM method can map T1, T2*, PD, susceptibility, and B1 simultaneously within a scan time that can be applied to human subjects.