ORIGINAL RESEARCH
Biexponential Diffusion Alterations in the Normal-Appearing White Matter of Glioma Patients Might Indicate the Presence of Global Vasogenic Edema Andrea Horv ath, MD,1,2 G abor Perlaki, PhD,1,3 Arnold T oth, MD,1,2,4 Gergely Orsi, PhD,1,3 Szilvia Nagy, MSc,1,5 Tam as D oczi, MD, DSc,2,3 Zsolt Horv ath, MD, PhD,2 and P eter Bogner, MD, PhD3* Purpose: To investigate normal-appearing white matter (NAWM) microstructure of glioma patients with biexponential diffusion analysis in order to reveal the nature of diffusion abnormalities and to assess whether they are region-specific or global. Materials and Methods: Twenty-four newly diagnosed glioma patients (grade II–IV) and 24 matched control subjects underwent diffusion-weighted imaging at 3T. Diffusion parameters were calculated using monoexponential and biexponential models. Apparent diffusion coefficient (ADC) values were measured in the entire NAWM of the hemisphere contralateral and ipsilateral to the tumor. In the contralateral NAWM, regional ADC values were assessed in the frontal, parietal, occipital, and temporal NAWM. Results: ADCmono and ADCfast were significantly higher than control values in all investigated regions except the temporal NAWM (P < 0.04). ADCslow was significantly increased in the total contralateral, frontal, and parietal NAWM (P < 0.03), while pslow was decreased in both total hemispheric NAWM and the parietal NAWM of glioma patients compared to controls (P < 0.04). ADCmono, ADCfast, ADCslow, and pslow were significantly different among the NAWM of the four lobes of the contralateral hemisphere in both groups (P < 0.0001), and these regional differences were similar in patients and controls (P > 0.05). Hemispheric ADCmono and pslow differences were different between groups (P < 0.05). Conclusion: Globally altered diffusion parameters suggest the presence of global vasogenic edema in the NAWM of glioma patients, which is further supported by the finding that regional differences in patients follow those found in controls. Alternatively, some tumor infiltration might contribute to diffusion abnormalities in the NAWM, especially in the tumor-affected hemisphere. J. MAGN. RESON. IMAGING 2016;00:000–000.
D
espite the intense effort in improving the outcome for glioma patients, their prognosis still remains poor.1 In order to develop efficient therapies and to make appropriate treatment decisions, it is necessary to know as much as possible about the behavior of gliomas. Elevated mean diffusivity has been found in the normal-appearing white matter (NAWM) of glioma patients, which was also present in the hemisphere unaffected by the tumor.2–4 The cause of this abnormality is unknown; however, tumor infiltration is the most widely accepted explanation2,3 that could possibly
influence treatment decision. Surprisingly, elevated apparent diffusion coefficient (ADC) values have also been found in the contralateral NAWM of meningioma patients, where clearly no tumor infiltration is present, which implies the contribution of other factors.5 Water diffusion in the brain is affected by tissue microstructure, and thus it can be used to monitor changes accompanying various pathologies.6 Signal decay with high b factors can better be described with a biexponential function.6–8 This model assumes a fast and slow diffusion pool
View this article online at wileyonlinelibrary.com. DOI: 10.1002/jmri.25202 Received Dec 10, 2015, and in revised form Feb 1, 2016. Accepted for publication Feb 2, 2016. *Address reprint requests to: P.B., 13 Ifj us ag st. P ecs, Hungary, H-7623. E-mail: [email protected] ecs, University of P ecs, P ecs, Hungary; 2Department of Neurosurgery, University of P ecs, P ecs, Hungary; 3MTA - PTE, From the 1Diagnostic Center of P ecs, P ecs, Hungary; and 5MTA - PTE, Neurobiology of Clinical Neuroscience MR Research Group, P ecs, Hungary; 4Department of Radiology, University of P Stress Research Group, P ecs, Hungary
C 2016 Wiley Periodicals, Inc. V 1
Journal of Magnetic Resonance Imaging
with fast and slow diffusion coefficients, whose origin is yet unclear. There are theories based on compartmentation and geometrical aspects.9,10 However, the most reasonable explanation is that the slow diffusion pool corresponds to highly structured water layers bound to membrane surfaces and cytoskeleton (ie, hydration shell around proteins and macromolecules) and the fast diffusion pool originates from the remaining extra- and intracellular tissue water.6,7,11 Consequently, diffusion imaging with an extended b factor range enables more specific tissue characterization and differentiation, and can be beneficial to a more complete understanding of diseases. The goal of the present study was 1) to further clarify the nature of elevated ADC in NAWM of glioma patients with a more detailed biexponential diffusion analysis, and 2) to reveal if this diffusion alteration was region-specific or global.
Materials and Methods Subjects This prospective study was approved by the Regional Research Ethics Committee. All subjects gave written informed consent before the examination. Inclusion criteria were newly diagnosed, unilateral grade II– IV gliomas with magnetic resonance imaging (MRI) scans before any intervention or treatment (surgery, chemoradiation) and age older than 18 years. Thirty patients were initially enrolled in the study. Three patients were excluded because white matter contralateral to the tumor showed hyperintense signal abnormalities on T2weighted scans and three patients were excluded due to unavailable histology. After exclusions, 24 patients with newly diagnosed, histologically verified glioma remained in the study. Patient characteristics are described in Table 1. As a control group, 24 age- and gender-matched healthy control subjects (14 males, 10 females, mean age 6 standard deviation [SD]: 42.08 6 10.94 years, age range: 23–63 years) were included.
MRI MRI was performed with a 3T Siemens TIM Trio MRI scanner (Siemens, Erlangen, Germany) with a 12-channel head coil. Conventional anatomical imaging included T1-, T2-weighted, and fluid-attenuated inversion recovery images. Three-dimensional T1-weighted magnetization-prepared rapid gradient-echo (MPRAGE) images (TR/TI/TE: 2530/1100/ 3.37 msec, flip angle 78, 176 sagittal slices, slice thickness 1 mm, field of view 256 3 256 mm2, matrix 256 3 256, receiver bandwidth 200 Hz/pixel, GRAPPA 2) served as structural scans. For diffusion-weighted imaging, a 2D trace-weighted single shot echo planar imaging sequence (TR/TE: 4800/128 msec, slice thickness 3.5 mm, distance factor 30%, field of view: 188 3 250 mm2, matrix 144 3 192, number of acquisitions: 5, bvalues: 0, 500, 1000, 2000, 3000, 4000, 5000 s/mm2) was used. 2
TABLE 1. Patient Demographics, Tumor Histology, and Tumor Locations
Patient characteristics Mean age 6 SD, range (years)
42.29 6 11.44, 20-61
Gender distribution
14M, 10F
Histology Grade II (n 5 13)
DA: 4, OD: 7, OA: 1, GA: 1
Grade III (n 5 4)
AA: 2, AOA: 1, AOD: 1
Grade IV (n 5 7)
GBM: 7
Tumor affected lobes Frontal
11
Parietal
7
Occipital
6
Temporal
2
SD: standard deviation, M: male, F: female, DA: diffuse astrocytoma, OD: oligodendroglioma, OA: oligoastrocytoma, GA: gemistocytic astrocytoma, AA: anaplastic astrocytoma, AOA: anaplastic oligoastrocytoma, AOD: anaplastic oligodendroglioma, GMB: glioblastoma multiforme.
Image Analysis Image analysis was performed by an observer with 5 years of experience in image processing (A.H.) who was blinded to the results of histology. The entire white matter (WM) in the hemisphere contralateral and ipsilateral to the tumor was automatically segmented on MPRAGE images using Freesurfer software.12 Besides the entire contralateral and ipsilateral hemispheric WM masks, regional regions of interest (ROIs) were also created by automatically parcellating the contralateral WM into frontal, parietal, occipital, and temporal lobe regions in each subject. In control subjects the corresponding WM regions were segmented (as in the age- and sex-matched patients). MPRAGE images were then registered to diffusion-weighted scans (6 degrees-of-freedom linear fit, correlation ratio cost function and trilinear interpolation) using FSL FMRIB’s Linear Image Registration Tool (FLIRT).13 Then the matrix of the MPRAGEto-diffusion coregistration was used to align the segmented binary WM masks to the diffusion space (trilinear interpolation). The resulting total contralateral and ipsilateral hemispheric WM masks were thresholded using a 0.9 threshold to avoid partial volume effects and to minimize possible impacts of misalignment between WM masks and diffusion-weighted images. All ROIs (regional and total hemispheric masks) were manually corrected to exclude any non-WM structure especially in the inferior part of the temporal and frontal lobes, which are prone to susceptibility artifacts, or any WM hyperintensities visible on T2-weighted scans (ie, b 5 0 s/mm2). Attention was paid to avoid any tumor and edema containing regions especially in the ipsilateral WM ROIs. Volume 00, No. 00
Horv ath et al.: Altered Biexponential Diffusion in NAWM
FIGURE 1: Normal-appearing white matter (NAWM) regions of interest (ROIs). (A) Total hemispheric NAWM ROIs and (B) regional NAWM ROIs overlaid on a diffusion-weighted image acquired with b 5 0 s/mm2 of a low-grade glioma patient. Light blue: total contralateral hemispheric NAWM; copper: total ipsilateral hemispheric NAWM; green: frontal NAWM; magenta: parietal NAWM; blue: temporal NAWM; yellow: occipital NAWM.
The resulting ROIs were defined as NAWM masks. Figure 1 illustrates hemispheric and regional NAWM masks. Diffusion parameters were assessed in all ROIs of patients and control subjects. ADCmono values were calculated by fitting the monoexponential signal decay over the low b-value range (ie, b-values of 0, 500, and 1000 s/mm2):
sb 2b:ADC 5e s0
(1)
where Sb is the signal intensity in the presence of diffusion sensitization and S0 is the signal intensity in the absence of diffusion sensitization. ADC values were broken down into ADCfast and ADCslow by applying the biexponential fit in the whole b-value range (ie, 0–5000 s/mm2):
sb 5pfast e 2bADCfast 1pslow e 2bADCslow ; s0
(2)
where ADCfast and ADCslow are the ADC values, and pfast and pslow are the contributors to the signal of the fast and slow diffusing water compartments (pslow 5 1-pfast). Curve fitting was performed with MatLab software’s curve-fitting toolbox (MathWorks, Natick, MA) and an in-house program code using a nonlinear least squares fit with a trust-region algorithm.
Statistical Analysis Statistical analyses were performed using SPSS 20.0 software (IBM, Armonk, NY). Normality of data distribution was tested by Shapiro–Wilk statistics. Homogeneity of variance was assessed by Levene’s test. Both total hemispheric and regional diffusion values were compared between patient and control groups by Student’s t-test or two-tailed Mann–Whitney U-test according to the statistical distribution of the data. In order to control for the potential confoundMonth 2016
ing effects of age, each comparison was repeated by creating a multiple linear regression model including age as a covariate. Gliomas might spread into the mirror regions of the contralateral hemisphere through the corpus callosum.14 Therefore, to control for the potential confounding effect of tumor location, the multiple linear regression model included tumor location as an additional covariate besides age in the regional analysis; tumor location was handled as a binary variable (present or absent) in the analysis of each lobe. The assumptions of multiple linear regression were satisfied, as judged by testing for linearity, normality assumptions of the residues, outliers, independence of errors, homoscedasticity, and multicollinearity.15 Two-way mixed analysis of variance (ANOVA) was performed to evaluate whether regional differences in diffusion parameters were different between patients and controls. The same analysis was used to assess whether hemispheric differences in diffusion values are the same in patients and controls. The results were considered significant if P < 0.05.
Results Comparison of Patients With Controls HEMISPHERIC ANALYSIS. Table 2 and Fig. 2 present the mean diffusion values in both total hemispheric NAWM of both groups. In the contralateral hemispheric NAWM, ADCmono, ADCfast, and ADCslow were significantly higher (P < 0.0001, P 5 0.0071, and P 5 0.0255, respectively), while pslow was significantly lower (P 5 0.0061) in the patient group compared to controls. In the hemisphere ipsilateral to the tumor, ADCmono and ADCfast were significantly higher, and pslow was significantly lower (P 5 0.0013, P 5 0.0372, and P 5 0.0011, respectively) in the NAWM of 3
4
8.24 6 0.62 7.64 6 0.24 8.01 6 0.32 7.56 6 0.32 8.27 6 0.39 7.82 6 0.27 8.33 6 0.43 7.90 6 0.30 8.21 6 0.39 7.95 6 0.42
patients
controls
patients
controls
patients
controls
patients
controls
patients
controls
Temporal
Occipital 0.0859
b
b
b
b
0.0004
0.0002
0.0002
0.0013
11.97 6 0.67
12.16 6 0.60
12.20 6 0.64
12.77 6 0.69
12.11 6 0.44
12.52 6 0.56
11.63 6 0.62
12.24 6 0.61
12.25 6 0.28
12.67 6 0.71
12.18 6 0.31
12.52 6 0.50
< 0.0001a a
ADCfast
P value
0.3936
b
b
b
b
0.0056
0.0053
0.0306
0.0372
a
0.0071a
P value
0.92 6 0.17
1.00 6 0.18
1.11 6 0.15
1.20 6 0.20
1.10 6 0.14
1.20 6 0.16
0.92 6 0.12
1.06 6 0.11
0.97 6 0.08
0.99 6 0.09
0.95 6 0.07
1.00 6 0.09
ADCslow
0.1747
0.1130
b
b
b
b
0.0195
0.0026
0.9351
a
0.0255a
P value
25.80 6 1.57
24.88 6 1.69
27.70 6 1.69
27.03 6 1.90
28.07 6 1.44
26.61 6 1.75
27.04 6 1.86
26.90 6 1.87
29.46 6 1.71
26.70 6 2.82
29.47 6 1.47
28.11 6 1.93
pslow (%)
0.1482b
0.2391b
0.0312b
0.0645b
0.0011a
0.0061a
P value
SD: standard deviation, NAWM: normal appearing white matter, ADCmono: apparent diffusion coefficient, ADCfast: ADC of the fast diffusion component, ADCslow: ADC of the slow diffusion component, pslow: volume fraction of the slow diffusion component. Significant results are marked with bold font. a Multiple linear regression with adjustment for age. b Multiple linear regression with adjustment for age and tumor location.
Contralateral
Parietal
7.59 6 0.23
controls
Frontal
7.96 6 0.32
patients
Contralateral
Ipsilateral
ADCmono
Groups
Locations
TABLE 2. Mean Diffusion Values 6 SD (1024mm2/s) in the NAWM and Results of the Statistical Comparisons of Patient and Control Subjects
Journal of Magnetic Resonance Imaging
Volume 00, No. 00
Horv ath et al.: Altered Biexponential Diffusion in NAWM
FIGURE 2: Results of the comparisons of monoexponential ADCmono values (A), biexponential ADCfast (B), ADCslow (C), and pslow (D) in the NAWM of glioma patients and healthy controls. In the boxplots, whiskers are set at minimum and maximum, the horizontal line marks the median, 1 indicates the mean, whereas box indicates the interquartile range (25–75%). *P < 0.05, **P < 0.01, ***P < 0.001. Contralateral and ipsilateral indicate the total NAWM in the hemisphere contralateral and ipsilateral to the tumor.
glioma patients compared to control values. ADCslow in the total NAWM ipsilateral to the tumor did not show any difference between the two groups (P 5 0.9351). REGIONAL NAWM ANALYSIS. Table 2 and Fig. 3 pres-
ent the mean diffusion values of the contralateral NAWM regions in both groups. The regional analysis of the NAWM contralateral to the tumor revealed significantly increased ADCmono and ADCfast values in the frontal (P 5 0.0002 and P 5 0.0306, respectively), parietal (P 5 0.0002 and P 5 0.0053, respectively), and occipital (P 5 0.0004 and P 5 0.0056, respectively) NAWM regions of glioma patients compared to healthy subjects. ADCslow was significantly higher in the frontal (P 5 0.0026) and parietal (P 5 0.0195) lobes of glioma patients compared with healthy subjects, while pslow was significantly decreased in the contralateral parietal NAWM of glioma patients compared with controls (P 5 0.0312). In the occipital NAWM, ADCslow and pslow did not differ between patients and controls (P 5 0.1130 and P 5 0.2391, respectively). None of the diffusion parameters showed significant differences between patients and controls in the temporal NAWM (P > 0.05). Month 2016
Comparison of Diffusion Parameters Among Hemispheres and Regions HEMISPHERIC ANALYSIS. The results of hemispheric analysis with two-way mixed ANOVA are presented in Table 3. ADCmono and pslow in the NAWM was significantly different between both hemispheres (P 5 0.0076 and P 5 0.0003, respectively). ADCfast and ADCslow did not differ between the two hemispehere’s NAWM. The significant group*hemisphere interactions in ADCmono and pslow analyses (P 5 0.0407 and P 5 0.0005, respectively) indicate that the ADCmono and pslow differences between the NAWM of the two hemispheres change differently in the two groups. REGIONAL ANALYSIS. The results of regional analysis with two-way mixed ANOVA are presented in Table 4. In the regional analysis of the NAWM in the hemisphere contralateral to the tumor, a significant difference was observed in all four diffusion parameters among regions (P < 0.0001). ADCmono in the frontal NAWM was significantly lower than all other regions (
Biexponential Diffusion Alterations in the Normal-Appearing White Matter of Glioma Patients Might Indicate the Presence of Global Vasogenic Edema Andrea Horv ath, MD,1,2 G abor Perlaki, PhD,1,3 Arnold T oth, MD,1,2,4 Gergely Orsi, PhD,1,3 Szilvia Nagy, MSc,1,5 Tam as D oczi, MD, DSc,2,3 Zsolt Horv ath, MD, PhD,2 and P eter Bogner, MD, PhD3* Purpose: To investigate normal-appearing white matter (NAWM) microstructure of glioma patients with biexponential diffusion analysis in order to reveal the nature of diffusion abnormalities and to assess whether they are region-specific or global. Materials and Methods: Twenty-four newly diagnosed glioma patients (grade II–IV) and 24 matched control subjects underwent diffusion-weighted imaging at 3T. Diffusion parameters were calculated using monoexponential and biexponential models. Apparent diffusion coefficient (ADC) values were measured in the entire NAWM of the hemisphere contralateral and ipsilateral to the tumor. In the contralateral NAWM, regional ADC values were assessed in the frontal, parietal, occipital, and temporal NAWM. Results: ADCmono and ADCfast were significantly higher than control values in all investigated regions except the temporal NAWM (P < 0.04). ADCslow was significantly increased in the total contralateral, frontal, and parietal NAWM (P < 0.03), while pslow was decreased in both total hemispheric NAWM and the parietal NAWM of glioma patients compared to controls (P < 0.04). ADCmono, ADCfast, ADCslow, and pslow were significantly different among the NAWM of the four lobes of the contralateral hemisphere in both groups (P < 0.0001), and these regional differences were similar in patients and controls (P > 0.05). Hemispheric ADCmono and pslow differences were different between groups (P < 0.05). Conclusion: Globally altered diffusion parameters suggest the presence of global vasogenic edema in the NAWM of glioma patients, which is further supported by the finding that regional differences in patients follow those found in controls. Alternatively, some tumor infiltration might contribute to diffusion abnormalities in the NAWM, especially in the tumor-affected hemisphere. J. MAGN. RESON. IMAGING 2016;00:000–000.
D
espite the intense effort in improving the outcome for glioma patients, their prognosis still remains poor.1 In order to develop efficient therapies and to make appropriate treatment decisions, it is necessary to know as much as possible about the behavior of gliomas. Elevated mean diffusivity has been found in the normal-appearing white matter (NAWM) of glioma patients, which was also present in the hemisphere unaffected by the tumor.2–4 The cause of this abnormality is unknown; however, tumor infiltration is the most widely accepted explanation2,3 that could possibly
influence treatment decision. Surprisingly, elevated apparent diffusion coefficient (ADC) values have also been found in the contralateral NAWM of meningioma patients, where clearly no tumor infiltration is present, which implies the contribution of other factors.5 Water diffusion in the brain is affected by tissue microstructure, and thus it can be used to monitor changes accompanying various pathologies.6 Signal decay with high b factors can better be described with a biexponential function.6–8 This model assumes a fast and slow diffusion pool
View this article online at wileyonlinelibrary.com. DOI: 10.1002/jmri.25202 Received Dec 10, 2015, and in revised form Feb 1, 2016. Accepted for publication Feb 2, 2016. *Address reprint requests to: P.B., 13 Ifj us ag st. P ecs, Hungary, H-7623. E-mail: [email protected] ecs, University of P ecs, P ecs, Hungary; 2Department of Neurosurgery, University of P ecs, P ecs, Hungary; 3MTA - PTE, From the 1Diagnostic Center of P ecs, P ecs, Hungary; and 5MTA - PTE, Neurobiology of Clinical Neuroscience MR Research Group, P ecs, Hungary; 4Department of Radiology, University of P Stress Research Group, P ecs, Hungary
C 2016 Wiley Periodicals, Inc. V 1
Journal of Magnetic Resonance Imaging
with fast and slow diffusion coefficients, whose origin is yet unclear. There are theories based on compartmentation and geometrical aspects.9,10 However, the most reasonable explanation is that the slow diffusion pool corresponds to highly structured water layers bound to membrane surfaces and cytoskeleton (ie, hydration shell around proteins and macromolecules) and the fast diffusion pool originates from the remaining extra- and intracellular tissue water.6,7,11 Consequently, diffusion imaging with an extended b factor range enables more specific tissue characterization and differentiation, and can be beneficial to a more complete understanding of diseases. The goal of the present study was 1) to further clarify the nature of elevated ADC in NAWM of glioma patients with a more detailed biexponential diffusion analysis, and 2) to reveal if this diffusion alteration was region-specific or global.
Materials and Methods Subjects This prospective study was approved by the Regional Research Ethics Committee. All subjects gave written informed consent before the examination. Inclusion criteria were newly diagnosed, unilateral grade II– IV gliomas with magnetic resonance imaging (MRI) scans before any intervention or treatment (surgery, chemoradiation) and age older than 18 years. Thirty patients were initially enrolled in the study. Three patients were excluded because white matter contralateral to the tumor showed hyperintense signal abnormalities on T2weighted scans and three patients were excluded due to unavailable histology. After exclusions, 24 patients with newly diagnosed, histologically verified glioma remained in the study. Patient characteristics are described in Table 1. As a control group, 24 age- and gender-matched healthy control subjects (14 males, 10 females, mean age 6 standard deviation [SD]: 42.08 6 10.94 years, age range: 23–63 years) were included.
MRI MRI was performed with a 3T Siemens TIM Trio MRI scanner (Siemens, Erlangen, Germany) with a 12-channel head coil. Conventional anatomical imaging included T1-, T2-weighted, and fluid-attenuated inversion recovery images. Three-dimensional T1-weighted magnetization-prepared rapid gradient-echo (MPRAGE) images (TR/TI/TE: 2530/1100/ 3.37 msec, flip angle 78, 176 sagittal slices, slice thickness 1 mm, field of view 256 3 256 mm2, matrix 256 3 256, receiver bandwidth 200 Hz/pixel, GRAPPA 2) served as structural scans. For diffusion-weighted imaging, a 2D trace-weighted single shot echo planar imaging sequence (TR/TE: 4800/128 msec, slice thickness 3.5 mm, distance factor 30%, field of view: 188 3 250 mm2, matrix 144 3 192, number of acquisitions: 5, bvalues: 0, 500, 1000, 2000, 3000, 4000, 5000 s/mm2) was used. 2
TABLE 1. Patient Demographics, Tumor Histology, and Tumor Locations
Patient characteristics Mean age 6 SD, range (years)
42.29 6 11.44, 20-61
Gender distribution
14M, 10F
Histology Grade II (n 5 13)
DA: 4, OD: 7, OA: 1, GA: 1
Grade III (n 5 4)
AA: 2, AOA: 1, AOD: 1
Grade IV (n 5 7)
GBM: 7
Tumor affected lobes Frontal
11
Parietal
7
Occipital
6
Temporal
2
SD: standard deviation, M: male, F: female, DA: diffuse astrocytoma, OD: oligodendroglioma, OA: oligoastrocytoma, GA: gemistocytic astrocytoma, AA: anaplastic astrocytoma, AOA: anaplastic oligoastrocytoma, AOD: anaplastic oligodendroglioma, GMB: glioblastoma multiforme.
Image Analysis Image analysis was performed by an observer with 5 years of experience in image processing (A.H.) who was blinded to the results of histology. The entire white matter (WM) in the hemisphere contralateral and ipsilateral to the tumor was automatically segmented on MPRAGE images using Freesurfer software.12 Besides the entire contralateral and ipsilateral hemispheric WM masks, regional regions of interest (ROIs) were also created by automatically parcellating the contralateral WM into frontal, parietal, occipital, and temporal lobe regions in each subject. In control subjects the corresponding WM regions were segmented (as in the age- and sex-matched patients). MPRAGE images were then registered to diffusion-weighted scans (6 degrees-of-freedom linear fit, correlation ratio cost function and trilinear interpolation) using FSL FMRIB’s Linear Image Registration Tool (FLIRT).13 Then the matrix of the MPRAGEto-diffusion coregistration was used to align the segmented binary WM masks to the diffusion space (trilinear interpolation). The resulting total contralateral and ipsilateral hemispheric WM masks were thresholded using a 0.9 threshold to avoid partial volume effects and to minimize possible impacts of misalignment between WM masks and diffusion-weighted images. All ROIs (regional and total hemispheric masks) were manually corrected to exclude any non-WM structure especially in the inferior part of the temporal and frontal lobes, which are prone to susceptibility artifacts, or any WM hyperintensities visible on T2-weighted scans (ie, b 5 0 s/mm2). Attention was paid to avoid any tumor and edema containing regions especially in the ipsilateral WM ROIs. Volume 00, No. 00
Horv ath et al.: Altered Biexponential Diffusion in NAWM
FIGURE 1: Normal-appearing white matter (NAWM) regions of interest (ROIs). (A) Total hemispheric NAWM ROIs and (B) regional NAWM ROIs overlaid on a diffusion-weighted image acquired with b 5 0 s/mm2 of a low-grade glioma patient. Light blue: total contralateral hemispheric NAWM; copper: total ipsilateral hemispheric NAWM; green: frontal NAWM; magenta: parietal NAWM; blue: temporal NAWM; yellow: occipital NAWM.
The resulting ROIs were defined as NAWM masks. Figure 1 illustrates hemispheric and regional NAWM masks. Diffusion parameters were assessed in all ROIs of patients and control subjects. ADCmono values were calculated by fitting the monoexponential signal decay over the low b-value range (ie, b-values of 0, 500, and 1000 s/mm2):
sb 2b:ADC 5e s0
(1)
where Sb is the signal intensity in the presence of diffusion sensitization and S0 is the signal intensity in the absence of diffusion sensitization. ADC values were broken down into ADCfast and ADCslow by applying the biexponential fit in the whole b-value range (ie, 0–5000 s/mm2):
sb 5pfast e 2bADCfast 1pslow e 2bADCslow ; s0
(2)
where ADCfast and ADCslow are the ADC values, and pfast and pslow are the contributors to the signal of the fast and slow diffusing water compartments (pslow 5 1-pfast). Curve fitting was performed with MatLab software’s curve-fitting toolbox (MathWorks, Natick, MA) and an in-house program code using a nonlinear least squares fit with a trust-region algorithm.
Statistical Analysis Statistical analyses were performed using SPSS 20.0 software (IBM, Armonk, NY). Normality of data distribution was tested by Shapiro–Wilk statistics. Homogeneity of variance was assessed by Levene’s test. Both total hemispheric and regional diffusion values were compared between patient and control groups by Student’s t-test or two-tailed Mann–Whitney U-test according to the statistical distribution of the data. In order to control for the potential confoundMonth 2016
ing effects of age, each comparison was repeated by creating a multiple linear regression model including age as a covariate. Gliomas might spread into the mirror regions of the contralateral hemisphere through the corpus callosum.14 Therefore, to control for the potential confounding effect of tumor location, the multiple linear regression model included tumor location as an additional covariate besides age in the regional analysis; tumor location was handled as a binary variable (present or absent) in the analysis of each lobe. The assumptions of multiple linear regression were satisfied, as judged by testing for linearity, normality assumptions of the residues, outliers, independence of errors, homoscedasticity, and multicollinearity.15 Two-way mixed analysis of variance (ANOVA) was performed to evaluate whether regional differences in diffusion parameters were different between patients and controls. The same analysis was used to assess whether hemispheric differences in diffusion values are the same in patients and controls. The results were considered significant if P < 0.05.
Results Comparison of Patients With Controls HEMISPHERIC ANALYSIS. Table 2 and Fig. 2 present the mean diffusion values in both total hemispheric NAWM of both groups. In the contralateral hemispheric NAWM, ADCmono, ADCfast, and ADCslow were significantly higher (P < 0.0001, P 5 0.0071, and P 5 0.0255, respectively), while pslow was significantly lower (P 5 0.0061) in the patient group compared to controls. In the hemisphere ipsilateral to the tumor, ADCmono and ADCfast were significantly higher, and pslow was significantly lower (P 5 0.0013, P 5 0.0372, and P 5 0.0011, respectively) in the NAWM of 3
4
8.24 6 0.62 7.64 6 0.24 8.01 6 0.32 7.56 6 0.32 8.27 6 0.39 7.82 6 0.27 8.33 6 0.43 7.90 6 0.30 8.21 6 0.39 7.95 6 0.42
patients
controls
patients
controls
patients
controls
patients
controls
patients
controls
Temporal
Occipital 0.0859
b
b
b
b
0.0004
0.0002
0.0002
0.0013
11.97 6 0.67
12.16 6 0.60
12.20 6 0.64
12.77 6 0.69
12.11 6 0.44
12.52 6 0.56
11.63 6 0.62
12.24 6 0.61
12.25 6 0.28
12.67 6 0.71
12.18 6 0.31
12.52 6 0.50
< 0.0001a a
ADCfast
P value
0.3936
b
b
b
b
0.0056
0.0053
0.0306
0.0372
a
0.0071a
P value
0.92 6 0.17
1.00 6 0.18
1.11 6 0.15
1.20 6 0.20
1.10 6 0.14
1.20 6 0.16
0.92 6 0.12
1.06 6 0.11
0.97 6 0.08
0.99 6 0.09
0.95 6 0.07
1.00 6 0.09
ADCslow
0.1747
0.1130
b
b
b
b
0.0195
0.0026
0.9351
a
0.0255a
P value
25.80 6 1.57
24.88 6 1.69
27.70 6 1.69
27.03 6 1.90
28.07 6 1.44
26.61 6 1.75
27.04 6 1.86
26.90 6 1.87
29.46 6 1.71
26.70 6 2.82
29.47 6 1.47
28.11 6 1.93
pslow (%)
0.1482b
0.2391b
0.0312b
0.0645b
0.0011a
0.0061a
P value
SD: standard deviation, NAWM: normal appearing white matter, ADCmono: apparent diffusion coefficient, ADCfast: ADC of the fast diffusion component, ADCslow: ADC of the slow diffusion component, pslow: volume fraction of the slow diffusion component. Significant results are marked with bold font. a Multiple linear regression with adjustment for age. b Multiple linear regression with adjustment for age and tumor location.
Contralateral
Parietal
7.59 6 0.23
controls
Frontal
7.96 6 0.32
patients
Contralateral
Ipsilateral
ADCmono
Groups
Locations
TABLE 2. Mean Diffusion Values 6 SD (1024mm2/s) in the NAWM and Results of the Statistical Comparisons of Patient and Control Subjects
Journal of Magnetic Resonance Imaging
Volume 00, No. 00
Horv ath et al.: Altered Biexponential Diffusion in NAWM
FIGURE 2: Results of the comparisons of monoexponential ADCmono values (A), biexponential ADCfast (B), ADCslow (C), and pslow (D) in the NAWM of glioma patients and healthy controls. In the boxplots, whiskers are set at minimum and maximum, the horizontal line marks the median, 1 indicates the mean, whereas box indicates the interquartile range (25–75%). *P < 0.05, **P < 0.01, ***P < 0.001. Contralateral and ipsilateral indicate the total NAWM in the hemisphere contralateral and ipsilateral to the tumor.
glioma patients compared to control values. ADCslow in the total NAWM ipsilateral to the tumor did not show any difference between the two groups (P 5 0.9351). REGIONAL NAWM ANALYSIS. Table 2 and Fig. 3 pres-
ent the mean diffusion values of the contralateral NAWM regions in both groups. The regional analysis of the NAWM contralateral to the tumor revealed significantly increased ADCmono and ADCfast values in the frontal (P 5 0.0002 and P 5 0.0306, respectively), parietal (P 5 0.0002 and P 5 0.0053, respectively), and occipital (P 5 0.0004 and P 5 0.0056, respectively) NAWM regions of glioma patients compared to healthy subjects. ADCslow was significantly higher in the frontal (P 5 0.0026) and parietal (P 5 0.0195) lobes of glioma patients compared with healthy subjects, while pslow was significantly decreased in the contralateral parietal NAWM of glioma patients compared with controls (P 5 0.0312). In the occipital NAWM, ADCslow and pslow did not differ between patients and controls (P 5 0.1130 and P 5 0.2391, respectively). None of the diffusion parameters showed significant differences between patients and controls in the temporal NAWM (P > 0.05). Month 2016
Comparison of Diffusion Parameters Among Hemispheres and Regions HEMISPHERIC ANALYSIS. The results of hemispheric analysis with two-way mixed ANOVA are presented in Table 3. ADCmono and pslow in the NAWM was significantly different between both hemispheres (P 5 0.0076 and P 5 0.0003, respectively). ADCfast and ADCslow did not differ between the two hemispehere’s NAWM. The significant group*hemisphere interactions in ADCmono and pslow analyses (P 5 0.0407 and P 5 0.0005, respectively) indicate that the ADCmono and pslow differences between the NAWM of the two hemispheres change differently in the two groups. REGIONAL ANALYSIS. The results of regional analysis with two-way mixed ANOVA are presented in Table 4. In the regional analysis of the NAWM in the hemisphere contralateral to the tumor, a significant difference was observed in all four diffusion parameters among regions (P < 0.0001). ADCmono in the frontal NAWM was significantly lower than all other regions (
