Improving the understanding and performance of clinical MRI using tissue property filters and the central contrast theorem, MASDIR pulse sequences and synergistic contrast MRI.

This paper updates and extends three previous papers on tissue property filters (TP-filters), Multiplied, Added, Divided and/or Subtracted Inversion Recovery (MASTIR) pulse sequences and synergistic contrast MRI (scMRI). It does this by firstly adding the central contrast theorem (CCT) to TP-filters, secondly including division with MASTIR sequences to make them Multiplied, Added, Subtracted and/or Divided IR (MASDIR) sequences, and thirdly incorporating division into the image processing needed for scMR to increase synergistic T1 contrast. These updated concepts are then used to explain and improve contrast at tissue boundaries, as well as to develop imaging regimes to detect and monitor small changes to the brain over time and quantify T1. The CCT is in two parts: the first part states that contrast produced by each TP is the product of the change in TP multiplied by the TP sequence weighting which is the first partial derivative of the TP-filter. The second part states that the overall fractional contrast is the algebraic sum of the fractional contrasts produced by each of the TPs. Subtraction of two IR sequences alone about doubles contrast relative to a conventional single IR sequence. Division of this subtraction can amplify contrast 5-15 times compared with conventional IR sequences. Dividing sequences can be problematic in areas where the signal is zero but this is avoided by dividing the difference in signal of two magnitude reconstructed IR sequences by the sum of their signals. The basis for the production of high contrast, high spatial resolution boundaries at white-gray matter junctions, between cerebral cortex and cerebrospinal fluid (CSF) and at other sites with subtracted IR (SIR) and divided subtracted IR (dSIR) sequences is explained and examples are shown. A key concept is the tissue fraction f, which is the proportion of a tissue in a mixture of two tissues within a voxel. Contrast at boundaries is a function of the partial derivative of the TP-filter, the partial derivative of the relevant TP with respect to f, and the partial derivative of f with respect to distance, x. Location of tissue boundaries is important for segmentation and is helpful in determining if inversion times have been chosen correctly. In small change regimes, the high sensitivity to small changes in T1 provided by dSIR images, together with the high definition boundaries, afford mechanisms for detecting small changes due to contrast agents, disease, perfusion and other causes. 3D isotropic rigid body registration provides a technique for following these changes over time in serial studies. Images showing high lesion contrast, high definition tissue and fluid boundaries, and the detection of small changes are included. T1 maps can be created by linearly scaling dSIR images.

MRI chemical shift artifact produced by center-out radial sampling of k-space: a potential pitfall in clinical diagnosis.

BACKGROUND: Center-out radial sampling of k-space in magnetic resonance imaging employs a different direction for each readout. Off-resonance artifacts (including those produced by chemical shift between water and fat) found with this type of sampling are usually described as blurring, however more specific characterization of these artifacts can be ascertained from the fact that their point spread function is ring-shaped. This produces effects that differ from those seen with Cartesian sampling of k-space. Experiments were designed to demonstrate the origin of these artifacts and a volunteer was imaged to show them. METHODS: Two phantoms containing oil in a syringe and an annulus of oil surrounded by water were scanned with a range of bandwidths from 62.5 down to 4 kHz. In a human volunteer, head, pelvis and spine images were obtained with bandwidths of 62.5 and 4 kHz. RESULTS: The two phantoms showed displacement of the oil signal away from the center into the region of the surrounding water. The effect increased as the bandwidth was decreased. In the head of the volunteer, signal from fat in red bone marrow in the skull was displaced centrally and peripherally relative to water within the marrow, and appeared in the region between the skull and the brain, as well as in the surrounding scalp. Displacements of the former type simulated subdural hematomas. Displacement of perivesical fat signal centrally over the wall of the bladder simulated bladder tumor, and displacement of fat signal from red bone marrow in the lumbar spine to the intervertebral discs simulated their cartilaginous endplates. CONCLUSIONS: Center-out radial artifacts are important to recognize on clinical images since they may mimic anatomy and simulate pathology. The article shows how these artifacts originate, includes examples, and describes how the artifacts differ from Cartesian chemical shift artifacts.

Constraints in estimating the proton density fat fraction.

The study evaluates four physically motivated constraints in the estimation of the proton density fat fraction (PDFF). Least squares approaches were developed for constraining the parameters in PDFF quantification based on the physics of magnetic resonance imaging. These were smooth fieldmap, smooth initial phase, nonnegative proton density and moderate R2∗ values. The constraints were evaluated in terms of their influence on the bias and standard deviation of the estimated parameters using numerical simulations and in vivo data acquired at 0.35 T. Results show that unconstrained least squares estimation is noisy and biased and that constraints can be effective at reducing both the standard deviation and bias.

MR image reconstruction using deep learning: evaluation of network structure and loss functions.

BACKGROUND: To review and evaluate approaches to convolutional neural network (CNN) reconstruction for accelerated cardiac MR imaging in the real clinical context. METHODS: Two CNN architectures, Unet and residual network (Resnet) were evaluated using quantitative and qualitative assessment by radiologist. Four different loss functions were also considered: pixel-wise (L1 and L2), patch-wise structural dissimilarity (Dssim) and feature-wise (perceptual loss). The networks were evaluated using retrospectively and prospectively under-sampled cardiac MR data. RESULTS: Based on our assessments, we find that Resnet and Unet achieve similar image quality but that former requires only 100,000 parameters compared to 1.3 million parameters for the latter. The perceptual loss function performed significantly better than L1, L2 or Dssim loss functions as determined by the radiologist scores. CONCLUSIONS: CNN image reconstruction using Resnet yields comparable image quality to Unet with 10X the number of parameters. This has implications for training with significantly lower data requirements. Network training using the perceptual loss function was found to better agree with radiologist scoring compared to L1, L2 or Dssim loss functions.

Associations between histologic features of nonalcoholic fatty liver disease (NAFLD) and quantitative diffusion-weighted MRI measurements in adults.

PURPOSE: To investigate in adults the associations between histologic features of nonalcoholic fatty liver disease (NAFLD) and quantitative measures derived from diffusion-weighted imaging (DWI). MATERIALS AND METHODS: Eighty-nine adults undergoing standard-of-care liver biopsy for NAFLD were recruited for DWI. Biopsies were scored for histologic features of NAFLD. DWI was performed using b-values of 0, 100, and 500 s/mm(2) . Images were reconstructed using either conventional magnitude averaging (CMA) or a method to address bulk motion artifacts (Beta*LogNormal, BLN). The apparent diffusion coefficient (ADC) and the diffusivity (D) and perfusion fraction (F) of the intravoxel incoherent motion (IVIM) model were measured in the right hepatic lobe using both reconstructions. Associations between histologic features and DWI-derived measures were tested statistically with several methods including multiple linear regression. RESULTS: Using CMA and BLN reconstructions, respectively, the means (and ranges) were 1.7 (1.1-3.5) and 1.4 (1.0-3.2) × 10(-3) mm(2) /s for ADC, 1.1 (0.84-1.4) and 0.84 (0.53-1.1) × 10(-3) mm(2) /s for D, and 17 and 18 (2.3-35)% for F. For both reconstruction methods, D decreased with steatosis and F decreased with fibrosis (P < 0.05). ADC was not independently associated with any histologic feature. CONCLUSION: Steatosis and fibrosis have significant independent effects on D and F in adults undergoing biopsy for NAFLD.

Proton density water fraction as a biomarker of bone marrow cellularity: validation in ex vivo spine specimens.

The purpose of this study was to evaluate a magnetic resonance imaging (MRI) technique for quantifying the proton density water fraction (PDWF) as a biomarker of bone marrow cellularity. Thirty-six human bone marrow specimens from 18 donors were excised and subjected to different measurements of tissue composition: PDWF quantification using a multiple gradient echo MRI technique, three biochemical assays (triglyceride, total lipid and water content) and a histological assessment of cellularity. Results showed a strong correlation between PDWF and bone marrow cellularity from histology (r=0.72). A strong correlation was also found between PDWF and the biochemical assay of water content (r=0.76). These results suggest the PDWF is a predictor of bone marrow cellularity in tissues and can provide a non-invasive assessment of bone marrow changes in clinical patients undergoing radiotherapy.

Fat composition changes in bone marrow during chemotherapy and radiation therapy.

PURPOSE: To quantify changes in bone marrow fat fraction and determine associations with peripheral blood cell counts. METHODS AND MATERIALS: In this prospective study, 19 patients received either highly myelotoxic treatment (radiation therapy plus cisplatin, 5-fluorouracil mitomycin C [FU/MMC], or cisplatin/5-FU/cetuximab) or less myelotoxic treatment (capecitabine-radiation therapy or no concurrent chemotherapy). Patients underwent MR imaging and venipuncture at baseline, midtreatment, and posttreatment visits. We performed mixed effects modeling of the mean proton density fat fraction (PDFF[%]) by linear time, treatment, and vertebral column region (lumbar [L]4-sacral [S]2 vs thoracic [T]10-L3 vs cervical[C]3-T9), while controlling for cumulative mean dose and other confounders. Spearman rank correlations were performed by white blood cell (WBC) counts versus the differences in PDFF(%) before and after treatment. RESULTS: Cumulative mean dose was associated with a 0.43% per Gy (P=.004) increase in PDFF(%). In the highly myelotoxic group, we observed significant changes in PDFF(%) per visit within L4-S2 (10.1%, P<.001) and within T10-L3 (3.93%, P=.01), relative to the reference C3-T9. In the less myelotoxic group, we did not observe significant changes in PDFF(%) per visit according to region. Within L4-S2, we observed a significant difference between treatment groups in the change in PDFF(%) per visit (5.36%, P=.04). Rank correlations of the inverse log differences in WBC versus the differences in PDFF(%) overall and within T10-S2 ranged from 0.69 to 0.78 (P<.05). Rank correlations of the inverse log differences in absolute neutrophil counts versus the differences in PDFF(%) overall and within L4-S2 ranged from 0.79 to 0.81 (P<.05). CONCLUSIONS: Magnetic resonance imaging fat quantification is sensitive to marrow composition changes that result from chemoradiation therapy. These changes are associated with peripheral blood cell counts. This study supports a rationale for bone marrow-sparing treatment planning to reduce the risk of hematologic toxicity.

Nonalcoholic fatty liver disease: MR imaging of liver proton density fat fraction to assess hepatic steatosis.

PURPOSE: To evaluate the diagnostic performance of magnetic resonance (MR) imaging-estimated proton density fat fraction (PDFF) for assessing hepatic steatosis in nonalcoholic fatty liver disease (NAFLD) by using centrally scored histopathologic validation as the reference standard. MATERIALS AND METHODS: This prospectively designed, cross-sectional, internal review board-approved, HIPAA-compliant study was conducted in 77 patients who had NAFLD and liver biopsy. MR imaging-PDFF was estimated from magnitude-based low flip angle multiecho gradient-recalled echo images after T2* correction and multifrequency fat modeling. Histopathologic scoring was obtained by consensus of the Nonalcoholic Steatohepatitis (NASH) Clinical Research Network Pathology Committee. Spearman correlation, additivity and variance stabilization for regression for exploring the effect of a number of potential confounders, and receiver operating characteristic analyses were performed. RESULTS: Liver MR imaging-PDFF was systematically higher, with higher histologic steatosis grade (P < .001), and was significantly correlated with histologic steatosis grade (ρ = 0.69, P < .001). The correlation was not confounded by age, sex, lobular inflammation, hepatocellular ballooning, NASH diagnosis, fibrosis, or magnetic field strength (P = .65). Area under the receiver operating characteristic curves was 0.989 (95% confidence interval: 0.968, 1.000) for distinguishing patients with steatosis grade 0 (n = 5) from those with grade 1 or higher (n = 72), 0.825 (95% confidence interval: 0.734, 0.915) to distinguish those with grade 1 or lower (n = 31) from those with grade 2 or higher (n = 46), and 0.893 (95% confidence interval: 0.809, 0.977) to distinguish those with grade 2 or lower (n = 58) from those with grade 3 (n = 19). CONCLUSION: MR imaging-PDFF showed promise for assessment of hepatic steatosis grade in patients with NAFLD. For validation, further studies with larger sample sizes are needed.

Error model for reduction of cardiac and respiratory motion effects in quantitative liver DW-MRI.

Diffusion-weighted images of the liver exhibit signal dropout from cardiac and respiratory motion, particularly in the left lobe. These artifacts cause bias and variance in derived parameters that quantify intravoxel incoherent motion. Many models of diffusion have been proposed, but few separate attenuation from diffusion or perfusion from that of bulk motion. The error model proposed here (Beta*LogNormal) is intended to accomplish that separation by modeling stochastic attenuation from bulk motion as multiplication by a Beta-distributed random variate. Maximum likelihood estimation with this error model can be used to derive intravoxel incoherent motion parameters separate from signal dropout, and does not require a priori specification of parameters to do so. Liver intravoxel incoherent motion parameters were derived for six healthy subjects under this error model and compared with least-squares estimates. Least-squares estimates exhibited bias due to cardiac and respiratory gating and due to location within the liver. Bias from these factors was significantly reduced under the Beta*LogNormal model, as was within-organ parameter variance. Similar effects were appreciable in diffusivity maps in two patients with focal liver lesions. These results suggest that, relative to least-squares estimation, the Beta*LogNormal model accomplishes the intended reduction of bias and variance from bulk motion in liver diffusion imaging.

Prospective study of functional bone marrow-sparing intensity modulated radiation therapy with concurrent chemotherapy for pelvic malignancies.

PURPOSE: To test the hypothesis that intensity modulated radiation therapy (IMRT) can reduce radiation dose to functional bone marrow (BM) in patients with pelvic malignancies (phase IA) and estimate the clinical feasibility and acute toxicity associated with this technique (phase IB). METHODS AND MATERIALS: We enrolled 31 subjects (19 with gynecologic cancer and 12 with anal cancer) in an institutional review board-approved prospective trial (6 in the pilot study, 10 in phase IA, and 15 in phase IB). The mean age was 52 years; 8 of 31 patients (26%) were men. Twenty-one subjects completed (18)F-fluorodeoxyglucose (FDG)-positron emission tomography (PET)/computed tomography (CT) simulation and magnetic resonance imaging by use of quantitative IDEAL (IDEAL IQ; GE Healthcare, Waukesha, WI). The PET/CT and IDEAL IQ were registered, and BM subvolumes were segmented above the mean standardized uptake value and below the mean fat fraction within the pelvis and lumbar spine; their intersection was designated as functional BM for IMRT planning. Functional BM-sparing vs total BM-sparing IMRT plans were compared in 12 subjects; 10 were treated with functional BM-sparing pelvic IMRT per protocol. RESULTS: In gynecologic cancer patients, the mean functional BM V(10) (volume receiving ≥10 Gy) and V(20) (volume receiving ≥20 Gy) were 85% vs 94% (P<.0001) and 70% vs 82% (P<.0001), respectively, for functional BM-sparing IMRT vs total BM-sparing IMRT. In anal cancer patients, the corresponding values were 75% vs 77% (P=.06) and 62% vs 67% (P=.002), respectively. Of 10 subjects treated with functional BM-sparing pelvic IMRT, 3 (30%) had acute grade 3 hematologic toxicity or greater. CONCLUSIONS: IMRT can reduce dose to BM subregions identified by (18)F-fluorodeoxyglucose-PET/CT and IDEAL IQ. The efficacy of BM-sparing IMRT is being tested in a phase II trial.

Assessment of liver fat quantification in the presence of iron.

This study assesses the stability of magnetic resonance liver fat measurements against changes in T2* due to the presence of iron, which is a confound for accurate quantification. The liver T2* was experimentally shortened by intravenous infusion of a super paramagnetic iron oxide contrast agent. Low flip angle multiecho gradient echo sequences were performed before, during and after infusion. The liver fat fraction (FF) was calculated in co-localized regions-of-interest using T2* models that assumed no decay, monoexponential decay and biexponential decay. Results show that, when T2* was neglected, there was a strong underestimation of FF and with monoexponential decay there was a weak overestimation of FF. Curve-fitting using the biexponential decay was found to be problematic. The overestimation of FF may be due to remaining deficiencies in the model, although is unlikely to be important for clinical diagnosis of steatosis.

Effect of shot number on the calculated apparent diffusion coefficient in phantoms and in human liver in diffusion-weighted echo-planar imaging.

PURPOSE: To show the signal intensity varies with shot number in diffusion-weighted (DW) echo-planar imaging (EPI) and affects apparent diffusion coefficient (ADC) calculation. MATERIALS AND METHODS: This prospective study was performed on 35 adult patients and 20 volunteers. Measurements were made on a 3T scanner using a breathhold DW spin-echo EPI (SE EPI) sequence. Three protocols were used: A) eight consecutive shots at a fixed b-value of 0 seconds/mm(2) with TR = 1000 and 3000 msec; B) seven consecutive shots at b-values = 0, 1000, 750, 500, 250, 100, 0 seconds/mm(2) (in that order) with TR = 3500 msec; and C) seven consecutive shots (as in B) with TR = 1000, 1750, and 7000 msec. RESULTS: For protocol A, signal intensity decreased significantly from the first to second shot (P<0.0001) and thereafter remained constant. For protocol B, the ADC depended on which b = 0 seconds/mm(2) image was used. Using the first b = 0 seconds/mm(2), the mean ADC was 15% higher than using the second b = 0 seconds/mm(2) (P<0.0001). For protocol C, the difference between ADC using the first b = 0 seconds/mm(2) and the second b = 0 seconds/mm(2) decreased as the TR increased. CONCLUSION: The signal intensity can vary with shot number in SE EPI. For TR > or = 3000 msec, steady-state is attained after one shot. Using data acquired prior to steady-state confounds the calculation of ADC values.

Nonalcoholic fatty liver disease: diagnostic and fat-grading accuracy of low-flip-angle multiecho gradient-recalled-echo MR imaging at 1.5 T.

PURPOSE: To assess the accuracy of four fat quantification methods at low-flip-angle multiecho gradient-recalled-echo (GRE) magnetic resonance (MR) imaging in nonalcoholic fatty liver disease (NAFLD) by using MR spectroscopy as the reference standard. MATERIALS AND METHODS: In this institutional review board-approved, HIPAA-compliant prospective study, 110 subjects (29 with biopsy-confirmed NAFLD, 50 overweight and at risk for NAFLD, and 31 healthy volunteers) (mean age, 32.6 years +/- 15.6 [standard deviation]; range, 8-66 years) gave informed consent and underwent MR spectroscopy and GRE MR imaging of the liver. Spectroscopy involved a long repetition time (to suppress T1 effects) and multiple echo times (to estimate T2 effects); the reference fat fraction (FF) was calculated from T2-corrected fat and water spectral peak areas. Imaging involved a low flip angle (to suppress T1 effects) and multiple echo times (to estimate T2* effects); imaging FF was calculated by using four analysis methods of progressive complexity: dual echo, triple echo, multiecho, and multiinterference. All methods except dual echo corrected for T2* effects. The multiinterference method corrected for multiple spectral interference effects of fat. For each method, the accuracy for diagnosis of fatty liver, as defined with a spectroscopic threshold, was assessed by estimating sensitivity and specificity; fat-grading accuracy was assessed by comparing imaging and spectroscopic FF values by using linear regression. RESULTS: Dual-echo, triple-echo, multiecho, and multiinterference methods had a sensitivity of 0.817, 0.967, 0.950, and 0.983 and a specificity of 1.000, 0.880, 1.000, and 0.880, respectively. On the basis of regression slope and intercept, the multiinterference (slope, 0.98; intercept, 0.91%) method had high fat-grading accuracy without statistically significant error (P > .05). Dual-echo (slope, 0.98; intercept, -2.90%), triple-echo (slope, 0.94; intercept, 1.42%), and multiecho (slope, 0.85; intercept, -0.15%) methods had statistically significant error (P < .05). CONCLUSION: Relaxation- and interference-corrected fat quantification at low-flip-angle multiecho GRE MR imaging provides high diagnostic and fat-grading accuracy in NAFLD.

Fatty liver disease: MR imaging techniques for the detection and quantification of liver steatosis.

Fatty liver disease is the most common cause of chronic liver disease in the United States. Noninvasive detection and quantification of fat is becoming more and more important clinically, due in large part to the growing prevalence of nonalcoholic fatty liver disease. Steatosis, the accumulation of fat-containing vacuoles within hepatocytes, is a key histologic feature of fatty liver disease. Liver biopsy, the current standard of reference for the assessment of steatosis, is invasive, has sampling errors, and is not appropriate in some settings. Several magnetic resonance (MR) imaging-based techniques--including chemical shift imaging, frequency-selective imaging, and MR spectroscopy--are currently in clinical use for the detection and quantification of fat-water admixtures, with each technique having important advantages, disadvantages, and limitations. These techniques permit the breakdown of the net MR signal into fat and water signal components, allowing the quantification of fat in liver tissue, and are increasingly being used in the diagnosis, treatment, and follow-up of fatty liver disease.