Purpose: A general problem of machine-learning algorithms based on the convolutional neural network (CNN) technique is that the reason for the output judgement is unclear. The purpose of this study was to introduce a strategy that may facilitate better understanding of how and why a specific judgement was made by the algorithm. The strategy is to preprocess the input image data in different ways to highlight the most important aspects of the images for reaching the output judgement.
Materials and Methods: T2-weighted brain image series falling into two age-ranges were used. Classifying each series into one of the two age-ranges was the given task for the CNN model. The images from each series were preprocessed in five different ways to generate five different image sets: (1) subimages from the inner area of the brain, (2) subimages from the periphery of the brain, (3–5) subimages of brain parenchyma, gray matter area, and white matter area, respectively, extracted from the subimages of (2). The CNN model was trained and tested in five different ways using one of these image sets. The network architecture and all the parameters for training and testing remained unchanged.
Results: The judgement accuracy achieved by training was different when the image set used for training was different. Some of the differences was statistically significant. The judgement accuracy decreased significantly when either extra-parenchymal or gray matter area was removed from the periphery of the brain (P < 0.05).
Conclusion: The proposed strategy may help visualize what features of the images were important for the algorithm to reach correct judgement, helping humans to understand how and why a particular judgement was made by a CNN.
Purpose: Post-contrast liver magnetic resonance imaging is typically performed with breath-hold 3D gradient echo sequences. However, breath-holding for >10 s is difficult for some patients. In this study, we compared the quality of hepatobiliary phase (HBP) imaging without breath-holding using the prototype pulse sequences stack-of-stars liver acquisition with volume acceleration (LAVA) (LAVA Star) with or without navigator echoes (LAVA Starnavi+ and LAVA Starnavi−) and Cartesian LAVA with navigator echoes (Cartesian LAVAnavi+).
Methods: Seventy-two patients were included in this single-center, retrospective, cross-sectional study. HBP imaging using the three LAVA sequences (Cartesian LAVAnavi+, LAVA Starnavi−, and LAVA Starnavi+) without breath-holding was performed for all patients using a 3T magnetic resonance system. Two independent radiologists qualitatively analyzed (overall image quality, liver edge sharpness, hepatic vein clarity, streak artifacts, and respiratory motion/pulsation artifacts) HBP images taken by the three sequences using a five-point scale. Quantitative evaluations were also performed by calculating the liver-to-spleen, -lesion, and -portal vein (PV) signal intensity ratios. The results were compared between the three sequences using the Friedman test.
Results: LAVA Starnavi+ showed the best image quality and hepatic vein clarity (P < 0.0001). LAVA Starnavi− showed the lowest image quality (P < 0.0001–0.0106). LAVA Starnavi+ images showed fewer streak artifacts than LAVA Starnavi− images (P < 0.0001), while Cartesian LAVAnavi+ images showed no streak artifacts. Cartesian LAVAnavi+ images showed stronger respiratory motion/pulsation artifacts than the others (P < 0.0001). LAVA Starnavi− images showed the highest liver-to-spleen ratios (P < 0.0001–0.0005). Cartesian LAVAnavi+ images showed the lowest liver-to-lesion and -PV ratios (P < 0.0001–0.0108).
Conclusion: In terms of image quality, the combination of stack-of-stars acquisition and navigator echoes is the best for HBP imaging without breath-holding.
Purpose: To compare different q-space reconstruction methods for undersampled diffusion spectrum imaging data.
Materials and Methods: We compared the quality of three methods: Mean Apparent Propagator (MAP); Compressed Sensing using Identity (CSI) and Compressed Sensing using Dictionary (CSD) with simulated data and in vivo acquisitions. We used retrospective undersampling so that the fully sampled reconstruction could be used as ground truth. We used the normalized mean squared error (NMSE) and the Pearson’s correlation coefficient as reconstruction quality indices. Additionally, we evaluated two propagator-based diffusion indices: mean squared displacement and return to zero probability. We also did a visual analysis around the centrum semiovale.
Results: All methods had reconstruction errors below 5% with low undersampling factors and with a wide range of noise levels. However, the CSD method had at least 1–2% lower NMSE than the other reconstruction methods at higher noise levels. MAP was the second-best method when using a sufficiently high number of q-space samples. MAP reconstruction showed better propagator-based diffusion indices for in vivo acquisitions. With undersampling factors greater than 4, MAP and CSI have noticeably more reconstruction error than CSD.
Conclusion: Undersampled data were best reconstructed by means of CSD in simulations and in vivo. MAP was more accurate in the extraction of propagator-based indices, particularly for in vivo data.
Purpose: Recently, the use of 3D real inversion recovery (3D-real IR) imaging has been proposed for the evaluation of endolymphatic hydrops (EH). This method shows similar contrast between the endolymphatic and perilymphatic spaces and surrounding bone compared with the hybrid of reversed image of positive endolymph signal and native image of perilymph signal multiplied with heavily T2-weighted MR cisternography (HYDROPS-Mi2) image. We measured the volume of the endolymphatic space using 3D-real IR and HYDROPS-Mi2 images, and compared the measurements obtained with both techniques.
Methods: HYDROPS-Mi2 and 3D-real IR images were obtained for 30 ears from 15 patients with clinical suspicion of EH; imaging was performed 4 h after intravenous administration of a single dose of gadolinium-based contrast agent. We measured the volume of the endolymphatic space in the cochlea and vestibule by manually drawing the regions of interest. The correlation between endolymphatic volume determined from HYDROPS-Mi2 images and 3D-real IR images was calculated.
Results: There was a strong positive linear correlation between the cochlear and vestibular endolymphatic volume determined from HYDROPS-Mi2 and 3D-real IR images. The Spearman’s rank correlation coefficient (ρ) between the measurements obtained with both images was 0.805 (P < 0.001) for the cochlea and 0.826 (P < 0.001) for the vestibule.
Conclusion: The endolymphatic volume measured using 3D-real IR images strongly correlated with that measured using HYDROPS-Mi2 images. Thus, 3D-real IR imaging might be a suitable method for the measurement of endolymphatic volume.
Purpose: This study assessed the MRI findings of strangulated small bowel obstruction (SBO) and mesenteric venous occlusion (MVO) in a rabbit model using 3T MRI.
Materials and Methods: Twenty rabbits were included in this study. The strangulated SBO and MVO models were generated via surgical procedures in nine rabbits, and sham surgery was performed in two rabbits. The success of generating the models was confirmed via angiographic, macroscopic, and microscopic findings after the surgical procedure. MRI was performed before and 30 min after inducing mesenteric ischemia. T1-weighted images (T1WIs), T2-weighted images (T2WIs), and fat-suppressed T2WIs (FS-T2WIs) were obtained using the BLADE technique, and fat-suppressed T1WIs (FS-T1WIs) were obtained. The signal intensities of the affected bowel before and after the surgical procedures were visually categorized as high, iso, and low intense compared with the findings for the normal bowel wall on all sequences. Bowel wall thickness was measured, and the signal intensity ratio (SI ratio) was calculated using the signal intensities of the bowel wall and psoas muscle.
Results: Angiographic, macroscopic, and microscopic findings confirmed that all surgical procedures were successful. The ischemic bowel wall was thicker than the normal bowel. The bowel wall was thicker in the MVO model (3.17 ± 0.55 mm) than in the strangulated SBO model (2.26 ± 0.46 mm). The signal intensity and SI ratio of the bowel wall were significantly higher after the procedure than before the procedure on all sequences in both models. The mesentery adjacent to the ischemic bowel loop exhibited a high signal intensity in all animals on FS-T2WIs.
Conclusion: Non-contrast MRI can be used to evaluate mesenteric ischemia caused by strangulated SBO and MVO. FS-T2WIs represented the best modality for depicting the high signal intensity in the bowel wall and mesentery caused by ischemia.
Purpose: Identifying plaque components such as intraplaque hemorrhage, lipid rich necrosis, and calcification is important to evaluate vulnerability of carotid atherosclerotic plaque; however, conventional vessel wall MR imaging may fail to discriminate plaque components. We aimed to evaluate the components of plaques using quantitative susceptibility mapping (QSM), a newly developed post-processing technique to provide voxel-based quantitative susceptibilities.
Methods: Seven patients scheduled for carotid endarterectomy were enrolled. Magnitude and phase images of five-echo 3D fast low angle shot (FLASH) were obtained using a 3T MRI, and QSM was calculated from the phase images. Conventional carotid vessel wall images (black-blood T1-weighted images [T1WI], T2-weighted images [T2WI], proton-density weighted images [PDWI], and time-of-flight images [TOF]) were also obtained. Pathological findings including intraplaque hemorrhage, calcification, and lipid rich necrosis at the thickest plaque section were correlated with relative susceptibility values with respect to the sternocleidomastoid muscle on QSM. On conventional vessel wall images, the contrast–noise ratio (CNR) between the three components and sternocleidomastoid muscle was measured respectively. Wilcoxon signed-rank test analyses were performed to assess the relative susceptibility values and CNR.
Results: Pathologically, lipid rich necrosis was proved in all of seven cases, and intraplaque hemorrhage in five of seven cases. Mean relative susceptibility value of hemorrhage was higher than lipid rich necrosis unexceptionally (P = 0.0313). There were no significant differences between CNR of hemorrhage and lipid rich necrosis on all sequences. In all six cases with plaque calcification, susceptibility value of calcification was significantly lower than lipid rich necrosis unexceptionally (P = 0.0156). There were significant differences between CNRs of lipid rich necrosis and calcification on T1WI, PDWI, TOF (P < 0.05).
Conclusion: QSM of carotid plaque would provide a novel quantitative MRI contrast that enables reliable differentiation among intraplaque hemorrhage, lipid rich necrosis, and calcification, and be useful to identify vulnerable plaques.
Purpose: It has been reported that leakage of intravenously administered gadolinium-based contrast agents (IV-GBCAs) into the cerebrospinal fluid (CSF) from the cortical veins even in healthy subjects can be detected using a highly sensitive pulse sequence such as heavily T2-weighted 3D fluid-attenuated inversion recovery and 3D-real inversion recovery (IR). The purpose of this study was to evaluate the feasibility of MR fingerprinting to detect GBCA leakage from the cortical veins after IV-GBCA.
Materials: Fourteen patients with suspected endolymphatic hydrops (EH) who received a single dose of IV-GBCA (39–79 years old) were included. The real IR images as well as MR fingerprinting images were obtained at 4 h after IV-GBCA. T1 and T2 values were obtained using MR fingerprinting and analyzed in ROIs covering intense GBCA leakage, and non-leakage areas of the CSF as determined on real IR images. The scan time for real IR imaging was 10 min and that for MR fingerprinting was 41 s.
Results: The mean T1 value of the ROI in the area of GBCA leakage was 2422 ± 261 ms and that in the non-leakage area was 3851 ± 235 ms (P < 0.01). There was no overlap between the T1 values in the area of GBCA leakage and those in the non-leakage area.
The mean T2 value in the area of GBCA leakage was 319 ± 90 ms and that in the non-leakage area was 670 ± 166 ms (P < 0.01). There was some overlap between the T2 values in the area of GBCA leakage and those in the non-leakage area.
Conclusion: Leaked GBCA from the cortical veins into the surrounding CSF can be detected using MR fingerprinting obtained in <1 min.
Purpose: We proposed and developed a new microstrip transmission line radiofrequency (RF) coil for a positron emission tomography (PET) insert for MRI, which has low electrical interactions with PET shield boxes. We performed imaging experiments using a single-channel and a four-channel proposed RF coils for proof-of-concept.
Methods: A conventional microstrip coil consists of a microstrip conductor, a ground conductor, and a dielectric between the two conductors. We proposed a microstrip coil for the PET insert that replaced the conventional single-layer ground conductor with the RF shield of the PET insert. A dielectric material, which could otherwise attenuate gamma photons radiated from the PET imaging tracer, was not used. As proof-of-concept, we compared conventional and the proposed single-channel coils. To study multichannel performance, we further developed a four-channel proposed RF coil. Since the MRI system had a single-channel transmission port, an interfacing four-way RF power division circuit was designed. The coils were implemented as both RF transmitters and receivers in a cylindrical frame of diameter 150 mm. Coil bench performances were tested with a network analyzer (Rohde & Schwarz, Germany), and a homogeneous phantom study was conducted for gradient echo imaging and RF field (B1) mapping in a 3T clinical MRI system (Verio, Siemens, Erlangen, Germany).
Results: For all coils, the power reflection coefficient was below −30 dB, and the transmission coefficients in the four-channel configuration were near or below −20 dB. The comparative single-channel coil study showed good similarity between the conventional and proposed coils. The gradient echo image of the four-channel coil showed expected flashing image intensity near the coils and no phase distortion was visible. Transmit B1 field map resembled the image performance.
Conclusion: The proposed PET-microstrip coil performed similarly to the conventional microstrip transmission line coil and is promising for the development of a compact coil-PET system capable of simultaneous PET/MRI analysis with an existing MRI system.
The silent navigator technique utilizes a non-selective excitation and an appropriate respiratory waveform generation method is necessary for an accurate motion detection. We compared three methods for silent navigator waveform generation. The profile generation method with coil selection (prof-selection) resulted in a high cross correlation with bellows signals and a large respiration amplitude. The prof-selection method should be used for silent navigator waveform generation.
We investigated the usefulness of diffusion tensor imaging using single-shot turbo spin-echo sequence (TSE–DTI) in detecting the responsible nerve root by multipoint measurements of fractional anisotropy (FA) values. Five patients with bilateral lumbar spinal stenosis showing unilateral neurological symptoms were examined using TSE–DTI. In the spinal canal, FA values in the symptomatic side were lower than those in the asymptomatic side. TSE–DTI using multipoint measurements of FA values can differentiate the responsible lumbar nerve root.