HHS Public Access Author manuscript Author Manuscript
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2017 June 01. Published in final edited form as: AJNR Am J Neuroradiol. 2016 December ; 37(12): 2265–2272. doi:10.3174/ajnr.A4939.
Comparison of quantitative cerebral blood flow measurements performed by Bookend Dynamic Susceptibility Contrast and Arterial Spin Labeling MRI in Relapsing Remitting MultipleSclerosis
Author Manuscript
Robert M. D’Ortenzio, MSc*, Seyed-Parsa Hojjat, PhD*, Rita Vitorino, BSc, Charles Grady Cantrell, MSc, Liesly Lee, MD, Anthony Feinstein, PhD, Paul O’Connor, MD, Timothy Carroll, PhD, and Richard I. Aviv, MBChB Sunnybrook Health Sciences Centre, Department of Psychiatry (AF), Neurology (LL), and Medical Imaging (SPH, RV, RIA), 2075 Bayview Ave., Toronto, ON, M4N 3M5 University of Toronto (RMD, POC, LL, AF, RIA), 27 King's College Circle, Toronto, Ontario, M5S 1A1. Department of Biomedical Engineering (CGC, TJC) and Radiology (TJC), Northwestern University, Chicago, Illinois 60611, USA
Abstract Author Manuscript
Background and Purpose—Quantitative CBF usage as a biomarker for cognitive impairment and disease progression in MS is potentially a powerful tool for longitudinal patient monitoring. Dynamic Susceptibility contrast perfusion with Bookend T1-calibration (Bookend Technique) and pseudo-continuous Arterial Spin Labeling have recently been used for CBF quantification in relapsing-remitting MS. The non-invasive nature of pseudo-continuous Arterial Spin Labeling is advantageous over gadolinium-based techniques, but correlation between the techniques is not well established in the context of MS. Materials and Methods—We compared pseudo-continuous Arterial Spin Labeling CBF to the Bookend technique in a prospective cohort of 19 healthy controls, 19 relapsing-remitting MS subjects without cognitive impairment and 20 relapsing-remitting MS subjects with cognitive
Author Manuscript
Corresponding author: Seyed-Parsa Hojjat Sunnybrook Health Sciences Centre, 2075 Bayview Ave., Room AB-204, Toronto, ON, M4N 3M5, Phone: 416-480-6100 x89617, Fax: 416-480-5218, [email protected]. *These authors contributed equally to this work DISCLOSURES: Liesly Lee—UNRELATED: Consultancy: Genzyme Canada, Novartis Canada, Biogen Canada, Roche Canada, Comments: Advisory board member; Grants/Grants Pending: Genzyme Canada, Novartis Canada, Biogen Canada, Comments: Clinical trial participation; Payment for Development of Educational Presentations: Biogen Canada, Comments: Review article publication. Anthony Feinstein— UNRELATED: Expert Testimony: MS Society of Canada*; Payment for Lectures (including service on speakers bureaus): Teva Neuroscience, Merck-Serono, Novartis, Biogen-Idec. Timothy Carroll— UNRELATED: Grants/ Grants Pending: NIH, * Comments: I have research grants, not directly related to the subject matter of this work; Patents (planned, pending or issued): I have a number of US patents not related to this work. * Richard Aviv—RELATED: Grant: CIHR, * Comments: Peer reviewed grant held by research institute. *money paid to institution Statement of Conflict of Interest The authors have no conflict of interest to declare regarding this study.
D’Ortenzio et al.
Page 2
Author Manuscript
impairment on a voxel-wise and Brodmann regional basis. Linear Pearson correlation, SNR and coefficient of variation were quantified. Results—Voxel-wise paired T-tests revealed no significant CBF differences between techniques after normalization of global mean intensities. Highest Pearson correlations were observed in deep GM structures (average r = 0.71 basal ganglia, and r = 0.65 thalamus) but remained robust for cortical GM, WM and WML (average r = 0.51, 0.53, 0.54 respectively). Lower Pearson correlations were observed for CL (average r=0.23). Brodmann regional correlations were significant for all groups. All correlations were maintained in healthy controls and in RRMS disease. Highest SNR was present in Bookend perfusion while the highest coefficient of variation was present in WML.
Author Manuscript
Conclusion—Agreement between pseudo-continuous Arterial Spin Labeling and Bookend technique CBF measurements is demonstrated in healthy controls and relapsing-remitting MS patients
Introduction Multiple Sclerosis is the most common non-traumatic cause of neurologic disability in young and middle age adults1. In addition to physical disability, cognitive impairment is a significant contributor to functional disability in MS patients and increasingly recognized as an important contributor to quality of life2. Cerebral hypoperfusion is well described in MS, initially with PET and recently with MRI3–6. The etiology is likely multifactorial7; primary and secondary hypoperfusion, metabolic dysfunction and primary cerebrovascular abnormality are all posited8.
Author Manuscript Author Manuscript
Few studies have evaluated the role of deep gray matter or cortical perfusion in cognition. Recently, quantitative CBF perfusion imaging is shown to correlate strongly with cognitive impairment in relapsing-remitting multiple sclerosis (RRMS)9–11. A pseudo-continuous Arterial Spin Labelling (pCASL) study demonstrated a 7% reduction in cortical CBF in early RRMS compared to healthy controls (HC) with cortical CBF reduction associated with lower memory scores in the RRMS group9. Another pCASL study demonstrated focal cerebral hypoperfusion in functionally significant areas in cognitively impaired RRMS groups compared to HC and cognitively preserved RRMS groups10. Similar results were shown using the Bookend perfusion technique in RRMS and secondary progressive MS patients where hypoperfusion explained ~7–20% of cognitive impairment respectively after correction for confounding factors11, 12. Importantly hypoperfusion was present in the cognitively impaired compared to non-impaired RRMS patients in the absence of structural differences. Cortical CBV reduction in MS was independently associated with overall cognitive impairment and correlated highly with individual cognitive tests13. Data are also available from studies evaluating deep gray matter (GM) perfusion. Inglese reported a significant CBV reduction in deep GM perfusion in primary progressive MS (PPMS) compared to relapsing remitting MS (RRMS)14. Deep GM CBV correlated with Color-word interference inhibition switching test (D-KEFSIS)14. These results demonstrate the apparently independent role cortical hypoperfusion plays in cognition in MS suggesting utility as a biomarker of cortical disease severity.
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2017 June 01.
D’Ortenzio et al.
Page 3
Author Manuscript Author Manuscript
The Bookend perfusion technique, which uses unique ‘bookend’ calibration scans on either side of a conventional DSC sequence, is validated against H2[O15] PET and demonstrates high test re-test repeatability15. Importantly, the technique demonstrates reproducible CBF measurements in both HC and disease states, providing both GM and WM perfusion metrics and demonstrating high signal to noise by exploiting T2* effects of gadolinium16, 17. The acquisition and processing is vendor-neutral and can be implemented using sequences already available on most MRI scanners. Disadvantages include the need for gadolinium administration, with a potential for increased adverse effects stimulating a growing interest in non-contrast based perfusion sequences such as pCASL18. pCASL exploits magnetically polarized protons in water molecules in arterial blood as an endogenous tracer and therefore does not require contrast administration19, 20. Quantitative CBF is achieved with good success especially in GM of healthy individuals21–23. Disadvantages include the inherently low SNR and longer transit times associated with the WM rendering WM CBF signal unreliable24. Lastly, pCASL is mostly limited to larger research centers with limited routine clinical use compared to DSC sequences. Given the options available for quantitative imaging and the potential role for longitudinal cognitive impairment monitoring in MS via perfusion measurements, we sought to compare the pCASL and Bookend perfusion techniques in HC and RRMS-NI and RRMS-I to assess their correlation coefficient of variation (CV) and SNR in the normal and disease states. We hypothesized that good correlation would be seen between the two techniques in the context of an RRMS study cohort, but the SNR would be higher in bookend CBF measurements.
Materials and Methods Patients Cohort
Author Manuscript
This study was approved by the Sunnybrook Ethics Review Board, and informed consent was obtained for all patients. RRMS patients were prospectively recruited from two tertiary referral MS clinics by a senior neurologist with specialist practice in MS (20 years’ experience) over 1 year. The Montreal Cognitive Assessment screening25 was initially utilized to identify 20 RRMS-I patients followed by 19 age- and gender-matched RRMS-NI patients and 19 HC. Exclusion criteria included history of drug/alcohol abuse; MS disease activity or steroid use within the past 3 months; premorbid (i.e. pre-MS) psychiatric history; head injury with loss of consciousness; concurrent medical diseases (e.g. cerebrovascular disease); and contraindication to MRI/gadolinium (e.g. impaired renal function). Cognitive testing
Author Manuscript
Briefly, cognitive testing was performed using a comprehensive test validated for use in MS (Minimal Assessment of Cognitive Function In Multiple Sclerosis) testing processing speed, working and visuospatial memory (Paced Auditory Serial Addition Test (PASAT), Symbol Digit Modalities Test (SDMT)), executive function(D-KEFS Sorting Test (DST)), verbal fluency(California Verbal Learning Test - II (CVLT2), Brief Visuospatial Test - revised (BVMT-R)), visuospatial perception and spatial processing(Judgment of Line Orientation Test (JLO)) and verbal fluency (Controlled Oral Work Association Test (COWAT))26. Consistent with convention, patients scoring more than 1.5 standard deviations below
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2017 June 01.
D’Ortenzio et al.
Page 4
Author Manuscript
normative data on 2 or more tests were defined as cognitively impaired12. Expanded Disability Status Scale (EDSS) and Hospital Anxiety (HADS – A) and Depression (HADS – D) Scales were evaluated for each patient. Image acquisition
Author Manuscript Author Manuscript
MR images were obtained using a 3T scanner (Achieva, Philips Healthcare, Best NL) using an 8-channel phased array head coil receiver. Acquisitions included a volumetric T1 turbo field echo (TR/TE/flip angle: 9.5 ms/2.3 ms/ 12°; matrix size: 256 × 219; FOV: 24 cm; slice thickness: 1.2 mm) and a proton density/T2 (TR/TE/flip angle: 2500 ms/10.7 ms/90°; matrix size: 256 × 263; FOV: 23 cm; slice thickness: 3 mm). Bookend perfusion included an echo planar DSC: TR/TE/flip angle: 1633 ms/30 ms/60°; matrix size: 96 × 93; FOV: 22 cm; in plane voxel size 2.3 × 2.4 mm; no gap; bandwidth: 1260 Hz/pixel; sections: 24; slice thickness 4mm. following 10 mL of Gadobutrol (1mmol/mL) (Gadovist; Bayer, Toronto, Canada) injected at a rate of 5mL/s (Medrad-Spectris Solaris, Bayer) immediately followed by a 25 mL bolus of saline at 5mL/s. Sixty acquisitions were acquired at 1.6 second intervals, with Gadobutrol administration on the 5th sequence. Segmented inversion recovery Look-Locker EPI sequence was performed immediately before and after the DSC sequence: TR/TE/flip angle: 29 ms/14 ms/20°; matrix size: 128×126; FOV: 22 cm; 15 lines in k-space per acquisition; slice thickness: 4 mm; scan time: 73 sec; 60 time points. In order to facilitate longitudinal magnetization recovery, a 3 sec delay was used after the last imaging time point20. pCASL images were acquired similar to Shirzadi et al27 with parameters (TR/TE = 4000/9.7 msec; matrix size: 64 × 64 × 18; voxel size: 3 × 3 × 5 mm3; label offset: 80 mm; first slice post label delay: 1600 msec; label duration (middle slice): 1800 msec, scan duration of 248seconds (30 Tag control pair). Reference ASL for absolute quantification of CBF were acquired (TR/ TE: 10,000/9.7 ms, 64 × 64 × 18 matrix, resolution: 3 × 3 × 5 mm3, and scan duration of 40 seconds)28. Perfusion processing Bookend perfusion utilizes pre- and post-gadolinium WM T1 values relative to T1 changes in the blood pool to carefully model the effects of intravascular-to-extravascular water exchange and quantify qCBV in WM independent of an arterial input function17. CBV and CBF quantification account for compartmentalization effects, average brain density and hematocrit differences between large arteries and capillaries29. Deconvolution of tissue concentration-time curves by the arterial input function using singular value decomposition is used to calculate rCBF while the ratio of area under the curve of the tissue-concentrationtime curve and arterial input function represents the rCBV. Finally, CBF is calculated as the product of rCBF, qCBV and rCBV30.
Author Manuscript
pCASL quantitative CBF calculations are completed using the image processing pipeline developed by Shirzadi et al27. Briefly, masks are created by automatic segmentation of GM voxels from the T1 image using the FAST tool (FMRIB’s Automated Segmentation)31. T1 images with tissue masks are then registered to ASL coordinate space. CBF calculations are then optimized by determining the point of maximum GM detectability with only differences in control and labelled images contributing to this peak level included using an
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2017 June 01.
D’Ortenzio et al.
Page 5
Author Manuscript
aligning/sorting/discarding/refining pipeline27. Finally CBF is defined using the following equation:
Where ΔM represents the mean signal in the difference (reference – tag) image; M0 represents the equilibrium magnetization signal extracted from the reference scan; α = 0.85 represents labeling efficiency; T1,b = 1.68s represents the longitudinal recovery time of arterial blood; Post Labelling Delay=1.6s; Δtz = acquisition time each axial slice; TE represents the echo time of the reference scan and relaxation time27.
represents the transverse tissue
Author Manuscript Author Manuscript Author Manuscript
Image processing—The reference pCASL control and pre-gadolinium DSC image acquisition were normalized to MNI152 space (Montreal Neurological Institute, McGill University) in SPM 8 (SPM8, Welcome Trust, London, UK) creating two study specific MNI152 templates. Each individual’s averaged pCASL control and pre-gadolinium DSC images were then registered to their study specific MNI template using a linear registration (FSL-FLIRT) followed by non-linear intensity modulation and multi-resolution non-linear registration with 4 subsampling levels (FSL-FNIRT). To better guide the alignment at each resolution level, the images were smoothed using full width half max Gaussian kernel (FWHM).. Basal ganglia (BG) and thalamus (TH) regions were manually traced for each subject by an experienced neuroradiologist (10 years’ experience) from T1-weighted images in native space. Similarly, WM, cortical lesions, and T1 black holes were manually traced with Analyze 8.0 (Mayo Clinic, Rochester, Minnestoa) on PD/T2 and Volumetric T1 respectively. These regions of interest were coregistered to MNI space as above and were removed from GM and WM regions but were independently analyzed. Resulting MNI images are 91 × 109 × 91 voxels, with each voxel measuring 2 × 2 × 2 mm. pCASL and Bookend modalities were correlated on a voxel-wise basis blinded to patient group, after constructing mean images by averaging each corresponding voxel’s CBF intensity over subjects within each cohort. A 7 × 7 voxel grid was created for each axial slice of the CBF maps and the mean CBF value within each grid section was calculated giving a wide range of CBF values32. To account for anatomical ROI variations occurring during average image construction, only non-zero CBF values were included in average calculations thereby ensuring that only overlapping voxels from each subject were considered. ROIs were subsegmented into GM and WM using subject specific masks generated using the Segment function in SPM8. GM was thresholded to have a minimum probability of 80% to reduce GM/WM overlap and partial volume effects. GM CBF intensities were also correlated on a Brodmann region basis. Brodmann areas were selected individually from average CBF maps for each subject group using a standard mask template in MNI152 space. GM CBF values were averaged over voxels contained within each Brodmann region, and average values were plotted.
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2017 June 01.
D’Ortenzio et al.
Page 6
Statistical Analysis
Author Manuscript
Baseline demographic data were compared between patients with and without impairment and HC using univariate general linear regression for age and logistic regression for gender. Age and gender were expressed as mean ± SD and proportions respectively. Paired t-tests were performed between pCASL and Bookend mean CBF intensities for both whole brain CBF maps to evaluate for statistically significant spatial intensity differences between sequences for each of GM, WM, BG and TH. In order to account for global overestimation of Bookend CBF32, mean intensities were normalized before paired t-test analysis. The Tvalue maps were thresholded at a false discovery rate corrected to p < 0.05, with a minimum cluster size of 20 voxels33.
Author Manuscript
Voxel-by voxel linear correlation was performed with the Pearson coefficient method using mean CBF intensity masked as WM, GM, BG, TH, WML and CL ROIs. To allow comparison of subject groups and ROIs, accounting for differences in voxel number, Fischer-Z scores were calculated and converted to standard-Z scores by dividing by the standard error ( ) where ν = number of voxels. Lastly Brodmann region, rather than voxel-wise, linear correlation was conducted using Pearson for each hemisphere. p< 0.05 was considered significant.
Author Manuscript
Dispersion between CBF imaging techniques was assessed using Bland – Altman techniques34. The coefficient of variation, a standard measure of dispersion (CV = (SD/μ) × 100) where μ is the mean CBF value across both modalities and SD represents the standard deviation of the difference of voxel-wise CBF values between imaging modalities, was calculated for mean CBF images in all subject groups. Noise was estimated as the mean of 100 randomly selected voxels outside of brain parenchyma from unsmoothed images. These voxels were plotted as a histogram of signal intensities and fitted with a Guassian distribution. To ensure the noise estimation had not been effected by processing, the standard deviation was compared to the mean of the noise distribution35. SNR was then defined as the ratio between mean noise estimation to mean GM CBF. All statistical analyses were completed using MATLAB (Mathworks Inc, Natick, Mass, US) software.
Results A summary of clinical data is presented in Table 1. There were no significant differences in age or gender between group cohorts (all p>0.05). Average age was 49.0 ± 7.1, 46.4 ± 7.2 and 48.1 ± 4.7 years for HC, RRMS-NI and RRMS-I groups respectively. Similarly, percent female gender was 60 % (12/20), 73.7 % (14/19) and 79 % (15/19) respectively.
Author Manuscript
Paired-T test comparison CBF measurements demonstrated greater global intensity using the pCASL technique compared to Bookend technique. No significant difference in regional voxel-by-voxel CBF signal intensities for any of WM, GM, BG and TH regions was present in paired-t tests between pCASL and Bookend perfusion for HC, RRMS-I, RRMS-NI groups after normalization of global mean intensities. Representative axial slices of both sequences are demonstrated in Figure 1.
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2017 June 01.
D’Ortenzio et al.
Page 7
Voxel-wise correlation
Author Manuscript
Results of mean CBF linear correlation are summarized in Table 2 and illustrated for HC in Figure 2a. There was no significant difference in correlation strengths between any groups. Although there is significant scatter in voxel-wise intensities, correlation r values are strong for HC, RRMS-NI and RRMS-I (average over ROIs r = 0.62, 0.64 and 0.59 respectively). Correlations were stronger for deep GM ROIs (average r = 0.71 BG, and r = 0.65 TH) compared to cortical GM,WM and WML regions (average r = 0.51, 0.53, 0.54 respectively). Correlation was weakest in CL regions (average r = 0.21). Slope values were all less than one (average = 0.51) indicating that in general pCASL derived CBF were of overall higher magnitude compared to Bookend CBF values. Coefficient of Variation and Signal to Noise Ratio
Author Manuscript
CV calculations demonstrate that on average, CL,WML followed by WM show greatest dispersion between Bookend and pCASL modalities (average CV =84.8%,66.8%, and 30.1% respectively) whereas the lowest dispersion was seen within the TH (average CV = 27.1%). Bland-Altman plots (Figure 2b) demonstrate good agreement of mean voxel-wise CBF values between perfusion imaging modalities. Bland-Altman plots of WM show mean differences significantly greater than zero in WM (13.7 ml/100g-min for HC). Estimated SNR was higher for Bookend measurements (mean SNR 6.6) compared to pCASL CBF measurements (mean SNR 4.3). Standard deviations (SD) of noise distributions are the same order of magnitude as the mean (mean noise pCASL ≈ 3.2 × SD, mean noise Bookend ≈ 2.9 × SD). CV and SNR values are presented in Table 3. GM Brodmann region based correlation
Author Manuscript
Results of Brodmann correlation are detailed in Table 4. Figure 3 shows a scatter plot of Brodmann region-wise analysis for the right hemisphere of the RRMS-I group. Each data point represents the average CBF within one Brodmann area. CBF intensities showed significant correlation for all test groups (at p
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2017 June 01. Published in final edited form as: AJNR Am J Neuroradiol. 2016 December ; 37(12): 2265–2272. doi:10.3174/ajnr.A4939.
Comparison of quantitative cerebral blood flow measurements performed by Bookend Dynamic Susceptibility Contrast and Arterial Spin Labeling MRI in Relapsing Remitting MultipleSclerosis
Author Manuscript
Robert M. D’Ortenzio, MSc*, Seyed-Parsa Hojjat, PhD*, Rita Vitorino, BSc, Charles Grady Cantrell, MSc, Liesly Lee, MD, Anthony Feinstein, PhD, Paul O’Connor, MD, Timothy Carroll, PhD, and Richard I. Aviv, MBChB Sunnybrook Health Sciences Centre, Department of Psychiatry (AF), Neurology (LL), and Medical Imaging (SPH, RV, RIA), 2075 Bayview Ave., Toronto, ON, M4N 3M5 University of Toronto (RMD, POC, LL, AF, RIA), 27 King's College Circle, Toronto, Ontario, M5S 1A1. Department of Biomedical Engineering (CGC, TJC) and Radiology (TJC), Northwestern University, Chicago, Illinois 60611, USA
Abstract Author Manuscript
Background and Purpose—Quantitative CBF usage as a biomarker for cognitive impairment and disease progression in MS is potentially a powerful tool for longitudinal patient monitoring. Dynamic Susceptibility contrast perfusion with Bookend T1-calibration (Bookend Technique) and pseudo-continuous Arterial Spin Labeling have recently been used for CBF quantification in relapsing-remitting MS. The non-invasive nature of pseudo-continuous Arterial Spin Labeling is advantageous over gadolinium-based techniques, but correlation between the techniques is not well established in the context of MS. Materials and Methods—We compared pseudo-continuous Arterial Spin Labeling CBF to the Bookend technique in a prospective cohort of 19 healthy controls, 19 relapsing-remitting MS subjects without cognitive impairment and 20 relapsing-remitting MS subjects with cognitive
Author Manuscript
Corresponding author: Seyed-Parsa Hojjat Sunnybrook Health Sciences Centre, 2075 Bayview Ave., Room AB-204, Toronto, ON, M4N 3M5, Phone: 416-480-6100 x89617, Fax: 416-480-5218, [email protected]. *These authors contributed equally to this work DISCLOSURES: Liesly Lee—UNRELATED: Consultancy: Genzyme Canada, Novartis Canada, Biogen Canada, Roche Canada, Comments: Advisory board member; Grants/Grants Pending: Genzyme Canada, Novartis Canada, Biogen Canada, Comments: Clinical trial participation; Payment for Development of Educational Presentations: Biogen Canada, Comments: Review article publication. Anthony Feinstein— UNRELATED: Expert Testimony: MS Society of Canada*; Payment for Lectures (including service on speakers bureaus): Teva Neuroscience, Merck-Serono, Novartis, Biogen-Idec. Timothy Carroll— UNRELATED: Grants/ Grants Pending: NIH, * Comments: I have research grants, not directly related to the subject matter of this work; Patents (planned, pending or issued): I have a number of US patents not related to this work. * Richard Aviv—RELATED: Grant: CIHR, * Comments: Peer reviewed grant held by research institute. *money paid to institution Statement of Conflict of Interest The authors have no conflict of interest to declare regarding this study.
D’Ortenzio et al.
Page 2
Author Manuscript
impairment on a voxel-wise and Brodmann regional basis. Linear Pearson correlation, SNR and coefficient of variation were quantified. Results—Voxel-wise paired T-tests revealed no significant CBF differences between techniques after normalization of global mean intensities. Highest Pearson correlations were observed in deep GM structures (average r = 0.71 basal ganglia, and r = 0.65 thalamus) but remained robust for cortical GM, WM and WML (average r = 0.51, 0.53, 0.54 respectively). Lower Pearson correlations were observed for CL (average r=0.23). Brodmann regional correlations were significant for all groups. All correlations were maintained in healthy controls and in RRMS disease. Highest SNR was present in Bookend perfusion while the highest coefficient of variation was present in WML.
Author Manuscript
Conclusion—Agreement between pseudo-continuous Arterial Spin Labeling and Bookend technique CBF measurements is demonstrated in healthy controls and relapsing-remitting MS patients
Introduction Multiple Sclerosis is the most common non-traumatic cause of neurologic disability in young and middle age adults1. In addition to physical disability, cognitive impairment is a significant contributor to functional disability in MS patients and increasingly recognized as an important contributor to quality of life2. Cerebral hypoperfusion is well described in MS, initially with PET and recently with MRI3–6. The etiology is likely multifactorial7; primary and secondary hypoperfusion, metabolic dysfunction and primary cerebrovascular abnormality are all posited8.
Author Manuscript Author Manuscript
Few studies have evaluated the role of deep gray matter or cortical perfusion in cognition. Recently, quantitative CBF perfusion imaging is shown to correlate strongly with cognitive impairment in relapsing-remitting multiple sclerosis (RRMS)9–11. A pseudo-continuous Arterial Spin Labelling (pCASL) study demonstrated a 7% reduction in cortical CBF in early RRMS compared to healthy controls (HC) with cortical CBF reduction associated with lower memory scores in the RRMS group9. Another pCASL study demonstrated focal cerebral hypoperfusion in functionally significant areas in cognitively impaired RRMS groups compared to HC and cognitively preserved RRMS groups10. Similar results were shown using the Bookend perfusion technique in RRMS and secondary progressive MS patients where hypoperfusion explained ~7–20% of cognitive impairment respectively after correction for confounding factors11, 12. Importantly hypoperfusion was present in the cognitively impaired compared to non-impaired RRMS patients in the absence of structural differences. Cortical CBV reduction in MS was independently associated with overall cognitive impairment and correlated highly with individual cognitive tests13. Data are also available from studies evaluating deep gray matter (GM) perfusion. Inglese reported a significant CBV reduction in deep GM perfusion in primary progressive MS (PPMS) compared to relapsing remitting MS (RRMS)14. Deep GM CBV correlated with Color-word interference inhibition switching test (D-KEFSIS)14. These results demonstrate the apparently independent role cortical hypoperfusion plays in cognition in MS suggesting utility as a biomarker of cortical disease severity.
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2017 June 01.
D’Ortenzio et al.
Page 3
Author Manuscript Author Manuscript
The Bookend perfusion technique, which uses unique ‘bookend’ calibration scans on either side of a conventional DSC sequence, is validated against H2[O15] PET and demonstrates high test re-test repeatability15. Importantly, the technique demonstrates reproducible CBF measurements in both HC and disease states, providing both GM and WM perfusion metrics and demonstrating high signal to noise by exploiting T2* effects of gadolinium16, 17. The acquisition and processing is vendor-neutral and can be implemented using sequences already available on most MRI scanners. Disadvantages include the need for gadolinium administration, with a potential for increased adverse effects stimulating a growing interest in non-contrast based perfusion sequences such as pCASL18. pCASL exploits magnetically polarized protons in water molecules in arterial blood as an endogenous tracer and therefore does not require contrast administration19, 20. Quantitative CBF is achieved with good success especially in GM of healthy individuals21–23. Disadvantages include the inherently low SNR and longer transit times associated with the WM rendering WM CBF signal unreliable24. Lastly, pCASL is mostly limited to larger research centers with limited routine clinical use compared to DSC sequences. Given the options available for quantitative imaging and the potential role for longitudinal cognitive impairment monitoring in MS via perfusion measurements, we sought to compare the pCASL and Bookend perfusion techniques in HC and RRMS-NI and RRMS-I to assess their correlation coefficient of variation (CV) and SNR in the normal and disease states. We hypothesized that good correlation would be seen between the two techniques in the context of an RRMS study cohort, but the SNR would be higher in bookend CBF measurements.
Materials and Methods Patients Cohort
Author Manuscript
This study was approved by the Sunnybrook Ethics Review Board, and informed consent was obtained for all patients. RRMS patients were prospectively recruited from two tertiary referral MS clinics by a senior neurologist with specialist practice in MS (20 years’ experience) over 1 year. The Montreal Cognitive Assessment screening25 was initially utilized to identify 20 RRMS-I patients followed by 19 age- and gender-matched RRMS-NI patients and 19 HC. Exclusion criteria included history of drug/alcohol abuse; MS disease activity or steroid use within the past 3 months; premorbid (i.e. pre-MS) psychiatric history; head injury with loss of consciousness; concurrent medical diseases (e.g. cerebrovascular disease); and contraindication to MRI/gadolinium (e.g. impaired renal function). Cognitive testing
Author Manuscript
Briefly, cognitive testing was performed using a comprehensive test validated for use in MS (Minimal Assessment of Cognitive Function In Multiple Sclerosis) testing processing speed, working and visuospatial memory (Paced Auditory Serial Addition Test (PASAT), Symbol Digit Modalities Test (SDMT)), executive function(D-KEFS Sorting Test (DST)), verbal fluency(California Verbal Learning Test - II (CVLT2), Brief Visuospatial Test - revised (BVMT-R)), visuospatial perception and spatial processing(Judgment of Line Orientation Test (JLO)) and verbal fluency (Controlled Oral Work Association Test (COWAT))26. Consistent with convention, patients scoring more than 1.5 standard deviations below
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2017 June 01.
D’Ortenzio et al.
Page 4
Author Manuscript
normative data on 2 or more tests were defined as cognitively impaired12. Expanded Disability Status Scale (EDSS) and Hospital Anxiety (HADS – A) and Depression (HADS – D) Scales were evaluated for each patient. Image acquisition
Author Manuscript Author Manuscript
MR images were obtained using a 3T scanner (Achieva, Philips Healthcare, Best NL) using an 8-channel phased array head coil receiver. Acquisitions included a volumetric T1 turbo field echo (TR/TE/flip angle: 9.5 ms/2.3 ms/ 12°; matrix size: 256 × 219; FOV: 24 cm; slice thickness: 1.2 mm) and a proton density/T2 (TR/TE/flip angle: 2500 ms/10.7 ms/90°; matrix size: 256 × 263; FOV: 23 cm; slice thickness: 3 mm). Bookend perfusion included an echo planar DSC: TR/TE/flip angle: 1633 ms/30 ms/60°; matrix size: 96 × 93; FOV: 22 cm; in plane voxel size 2.3 × 2.4 mm; no gap; bandwidth: 1260 Hz/pixel; sections: 24; slice thickness 4mm. following 10 mL of Gadobutrol (1mmol/mL) (Gadovist; Bayer, Toronto, Canada) injected at a rate of 5mL/s (Medrad-Spectris Solaris, Bayer) immediately followed by a 25 mL bolus of saline at 5mL/s. Sixty acquisitions were acquired at 1.6 second intervals, with Gadobutrol administration on the 5th sequence. Segmented inversion recovery Look-Locker EPI sequence was performed immediately before and after the DSC sequence: TR/TE/flip angle: 29 ms/14 ms/20°; matrix size: 128×126; FOV: 22 cm; 15 lines in k-space per acquisition; slice thickness: 4 mm; scan time: 73 sec; 60 time points. In order to facilitate longitudinal magnetization recovery, a 3 sec delay was used after the last imaging time point20. pCASL images were acquired similar to Shirzadi et al27 with parameters (TR/TE = 4000/9.7 msec; matrix size: 64 × 64 × 18; voxel size: 3 × 3 × 5 mm3; label offset: 80 mm; first slice post label delay: 1600 msec; label duration (middle slice): 1800 msec, scan duration of 248seconds (30 Tag control pair). Reference ASL for absolute quantification of CBF were acquired (TR/ TE: 10,000/9.7 ms, 64 × 64 × 18 matrix, resolution: 3 × 3 × 5 mm3, and scan duration of 40 seconds)28. Perfusion processing Bookend perfusion utilizes pre- and post-gadolinium WM T1 values relative to T1 changes in the blood pool to carefully model the effects of intravascular-to-extravascular water exchange and quantify qCBV in WM independent of an arterial input function17. CBV and CBF quantification account for compartmentalization effects, average brain density and hematocrit differences between large arteries and capillaries29. Deconvolution of tissue concentration-time curves by the arterial input function using singular value decomposition is used to calculate rCBF while the ratio of area under the curve of the tissue-concentrationtime curve and arterial input function represents the rCBV. Finally, CBF is calculated as the product of rCBF, qCBV and rCBV30.
Author Manuscript
pCASL quantitative CBF calculations are completed using the image processing pipeline developed by Shirzadi et al27. Briefly, masks are created by automatic segmentation of GM voxels from the T1 image using the FAST tool (FMRIB’s Automated Segmentation)31. T1 images with tissue masks are then registered to ASL coordinate space. CBF calculations are then optimized by determining the point of maximum GM detectability with only differences in control and labelled images contributing to this peak level included using an
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2017 June 01.
D’Ortenzio et al.
Page 5
Author Manuscript
aligning/sorting/discarding/refining pipeline27. Finally CBF is defined using the following equation:
Where ΔM represents the mean signal in the difference (reference – tag) image; M0 represents the equilibrium magnetization signal extracted from the reference scan; α = 0.85 represents labeling efficiency; T1,b = 1.68s represents the longitudinal recovery time of arterial blood; Post Labelling Delay=1.6s; Δtz = acquisition time each axial slice; TE represents the echo time of the reference scan and relaxation time27.
represents the transverse tissue
Author Manuscript Author Manuscript Author Manuscript
Image processing—The reference pCASL control and pre-gadolinium DSC image acquisition were normalized to MNI152 space (Montreal Neurological Institute, McGill University) in SPM 8 (SPM8, Welcome Trust, London, UK) creating two study specific MNI152 templates. Each individual’s averaged pCASL control and pre-gadolinium DSC images were then registered to their study specific MNI template using a linear registration (FSL-FLIRT) followed by non-linear intensity modulation and multi-resolution non-linear registration with 4 subsampling levels (FSL-FNIRT). To better guide the alignment at each resolution level, the images were smoothed using full width half max Gaussian kernel (FWHM).. Basal ganglia (BG) and thalamus (TH) regions were manually traced for each subject by an experienced neuroradiologist (10 years’ experience) from T1-weighted images in native space. Similarly, WM, cortical lesions, and T1 black holes were manually traced with Analyze 8.0 (Mayo Clinic, Rochester, Minnestoa) on PD/T2 and Volumetric T1 respectively. These regions of interest were coregistered to MNI space as above and were removed from GM and WM regions but were independently analyzed. Resulting MNI images are 91 × 109 × 91 voxels, with each voxel measuring 2 × 2 × 2 mm. pCASL and Bookend modalities were correlated on a voxel-wise basis blinded to patient group, after constructing mean images by averaging each corresponding voxel’s CBF intensity over subjects within each cohort. A 7 × 7 voxel grid was created for each axial slice of the CBF maps and the mean CBF value within each grid section was calculated giving a wide range of CBF values32. To account for anatomical ROI variations occurring during average image construction, only non-zero CBF values were included in average calculations thereby ensuring that only overlapping voxels from each subject were considered. ROIs were subsegmented into GM and WM using subject specific masks generated using the Segment function in SPM8. GM was thresholded to have a minimum probability of 80% to reduce GM/WM overlap and partial volume effects. GM CBF intensities were also correlated on a Brodmann region basis. Brodmann areas were selected individually from average CBF maps for each subject group using a standard mask template in MNI152 space. GM CBF values were averaged over voxels contained within each Brodmann region, and average values were plotted.
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2017 June 01.
D’Ortenzio et al.
Page 6
Statistical Analysis
Author Manuscript
Baseline demographic data were compared between patients with and without impairment and HC using univariate general linear regression for age and logistic regression for gender. Age and gender were expressed as mean ± SD and proportions respectively. Paired t-tests were performed between pCASL and Bookend mean CBF intensities for both whole brain CBF maps to evaluate for statistically significant spatial intensity differences between sequences for each of GM, WM, BG and TH. In order to account for global overestimation of Bookend CBF32, mean intensities were normalized before paired t-test analysis. The Tvalue maps were thresholded at a false discovery rate corrected to p < 0.05, with a minimum cluster size of 20 voxels33.
Author Manuscript
Voxel-by voxel linear correlation was performed with the Pearson coefficient method using mean CBF intensity masked as WM, GM, BG, TH, WML and CL ROIs. To allow comparison of subject groups and ROIs, accounting for differences in voxel number, Fischer-Z scores were calculated and converted to standard-Z scores by dividing by the standard error ( ) where ν = number of voxels. Lastly Brodmann region, rather than voxel-wise, linear correlation was conducted using Pearson for each hemisphere. p< 0.05 was considered significant.
Author Manuscript
Dispersion between CBF imaging techniques was assessed using Bland – Altman techniques34. The coefficient of variation, a standard measure of dispersion (CV = (SD/μ) × 100) where μ is the mean CBF value across both modalities and SD represents the standard deviation of the difference of voxel-wise CBF values between imaging modalities, was calculated for mean CBF images in all subject groups. Noise was estimated as the mean of 100 randomly selected voxels outside of brain parenchyma from unsmoothed images. These voxels were plotted as a histogram of signal intensities and fitted with a Guassian distribution. To ensure the noise estimation had not been effected by processing, the standard deviation was compared to the mean of the noise distribution35. SNR was then defined as the ratio between mean noise estimation to mean GM CBF. All statistical analyses were completed using MATLAB (Mathworks Inc, Natick, Mass, US) software.
Results A summary of clinical data is presented in Table 1. There were no significant differences in age or gender between group cohorts (all p>0.05). Average age was 49.0 ± 7.1, 46.4 ± 7.2 and 48.1 ± 4.7 years for HC, RRMS-NI and RRMS-I groups respectively. Similarly, percent female gender was 60 % (12/20), 73.7 % (14/19) and 79 % (15/19) respectively.
Author Manuscript
Paired-T test comparison CBF measurements demonstrated greater global intensity using the pCASL technique compared to Bookend technique. No significant difference in regional voxel-by-voxel CBF signal intensities for any of WM, GM, BG and TH regions was present in paired-t tests between pCASL and Bookend perfusion for HC, RRMS-I, RRMS-NI groups after normalization of global mean intensities. Representative axial slices of both sequences are demonstrated in Figure 1.
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2017 June 01.
D’Ortenzio et al.
Page 7
Voxel-wise correlation
Author Manuscript
Results of mean CBF linear correlation are summarized in Table 2 and illustrated for HC in Figure 2a. There was no significant difference in correlation strengths between any groups. Although there is significant scatter in voxel-wise intensities, correlation r values are strong for HC, RRMS-NI and RRMS-I (average over ROIs r = 0.62, 0.64 and 0.59 respectively). Correlations were stronger for deep GM ROIs (average r = 0.71 BG, and r = 0.65 TH) compared to cortical GM,WM and WML regions (average r = 0.51, 0.53, 0.54 respectively). Correlation was weakest in CL regions (average r = 0.21). Slope values were all less than one (average = 0.51) indicating that in general pCASL derived CBF were of overall higher magnitude compared to Bookend CBF values. Coefficient of Variation and Signal to Noise Ratio
Author Manuscript
CV calculations demonstrate that on average, CL,WML followed by WM show greatest dispersion between Bookend and pCASL modalities (average CV =84.8%,66.8%, and 30.1% respectively) whereas the lowest dispersion was seen within the TH (average CV = 27.1%). Bland-Altman plots (Figure 2b) demonstrate good agreement of mean voxel-wise CBF values between perfusion imaging modalities. Bland-Altman plots of WM show mean differences significantly greater than zero in WM (13.7 ml/100g-min for HC). Estimated SNR was higher for Bookend measurements (mean SNR 6.6) compared to pCASL CBF measurements (mean SNR 4.3). Standard deviations (SD) of noise distributions are the same order of magnitude as the mean (mean noise pCASL ≈ 3.2 × SD, mean noise Bookend ≈ 2.9 × SD). CV and SNR values are presented in Table 3. GM Brodmann region based correlation
Author Manuscript
Results of Brodmann correlation are detailed in Table 4. Figure 3 shows a scatter plot of Brodmann region-wise analysis for the right hemisphere of the RRMS-I group. Each data point represents the average CBF within one Brodmann area. CBF intensities showed significant correlation for all test groups (at p
