Serial Susceptibility Weighted MRI measures brain iron and microbleeds in dementia

2.1 Participant selection

Over a 5-year period, 1,348 individuals from several local communities were screened to recruit elderly cognitively normal and mild cognitively impaired (MCI) study participants. Inclusion and exclusion criteria defined by the Mayo Clinic Group [23] were used in the screening process. All Normals were without objective or subjective memory deficits and within normal limits on neuropsychological testing (Global Clinical Dementia Rating (CDR) of 0, CDR memory component of 0 and a sum of CDR boxes of 1 or less at baseline).[24] All MCI participants fulfilled the Mayo Clinic criteria for classification as MCI-multiple domain impairment, i.e.: i) A memory complaint confirmed by Logical Memory testing or reports of the informant and a CDR of 0.5. ii) Normal activities of daily living. iii) Normal general cognitive function iv) Abnormal memory for age as measured by standard scores and education. v) A global CDR of 0.5 and no clinically determined dementia.

Participants in the study have been evaluated cognitively (bi-yearly) and radiologically (every 9–14 months) for up to 5 years. Cognitive tests were scheduled in proximity to MR evaluations. All participants gave informed consent and all studies were approved by the Loma Linda University Institutional Review Board.

2.2 Cognitive testing

Cognitive tests included: videotaped CDR with informant, Logical Memory I, II, Word Fluency: Phonemic and Semantic, Wisconsin Card Sorting Test, Trail Making Test A&B, Boston Naming Test, Draw-A-Clock, and Geriatric Depression Scale. North American Adult Reading Test and Depression Features Battery Version II were given during initial assessment. Results of imaging and cognitive studies were reviewed bimonthly and, if neurologic assessment indicated development of a disorder other than AD, the subject was removed from the study. Results of neuropsychological tests were considered abnormal if below 1.5 standard deviations (SD) on normative data based on age and education. The diagnosis of dementia was based on clinical judgment, a consensus conference, NINCDS-ADRDA criteria, and a Sum of Boxes (SOB) on a CDR ≥ 3.5.[24]

2.3 MR imaging

MR acquisitions were performed on a 1.5T MR scanner (Vision, Siemens Medical Solutions) and included: sagittal magnetization prepared rapid gradient echo (T1-3D MPRAGE), axial T2 weighted fast spin echo (T2-FSE), fluid attenuated inversion recovery (FLAIR), 2D T2* weighted gradient recalled echo acquisition (T2* GRE) and 3D whole brain SWI.

SWI is a fully flow-compensated, three-dimensional (3D), high-resolution, gradient-echo sequence used to collect magnitude and phase data with each slice. To obtain SWI phase measurements, phase images were high-pass-filtered and processed region by region offline with a custom Visual C++ software program (SPIN, Signal Processing in NMR).[25] This software was also used to obtain SWI images, by multiplying magnitude images with filtered phase images to enhance the susceptibility effect and then performing a minimum intensity projection (mIP) reconstruction to produce various slice thicknesses (4 – 10 mm) in order to better visualize connectivity of the microvasculature to BMB. Imaging acquisition parameters for SWI used in-plane resolution 0.5 mm × 1.0 mm (interpolated to 0.5 mm × 0.5 mm); 2 mm thick, field-of-view = 256 mm × 256 mm; matrix = 256 × 512, 48 slices; TE = 40 ms; TR = 57 ms; and FA = 15 degrees. Phase units were expressed as Siemens normalized units as discussed previously.[7] Imaging parameters for the 2D T2* GRE sequence used for comparison to SWI were in-plane resolution 0.5 mm × 1.0 mm (interpolated to 0.5 mm × 0.5 mm); 4 mm thick, field-of-view = 256 mm × 256 mm; matrix = 256 × 512, 24 slices; TE = 18 ms; TR = 500 ms; and FA = 15 degrees.

2.4 ROI regional tissue iron determinations

Phase images from whole brain SWI were used to quantitatively measure iron in 14 different regions of interest (ROIs) in both cerebral hemispheres. Methods for drawing ROIs in each cerebral hemisphere, post-processing for phase measurements, and baseline phase data for the ROIs have been given in previous publications.[7, 22, 26] Bilateral ROIs assayed included: motor cortex grey matter (GM), cerebral white matter subjacent to motor cortex (WM), cerebrospinal fluid in the sulcus next to the motor cortex (CSF), caudate nucleus (CN), paracingulate gyrus (PG), thalamus (T), total putamen (P), globus pallidus (GP), red nucleus (RN), (RN-nonvascular), (RN-vascular), substantia nigra (SN), (SN-pars reticularis), (SN-pars compacta). A demonstration of ROI placement for serial putaminal studies is shown in Figure 1. Phase measurements obtained from SWI filtered phase data were described in normalized units as supplied by the manufacturer and were inversely associated with putative iron values. As demonstrated previously in phantoms, measurements in an ROI of at least 100 pixels enabled a determination of changing iron levels of 3 µgm Fe/gm brain (p=0.01).[7] To check reproducibility of phase measurements, studies were processed at both participating sites with excellent interobserver agreement (intraclass coefficient (ICC) of 0.94 determined on 43 test cases).[7]

2.5 Brain microbleed (BMB) identification

BMB were defined for this study as homogeneous signal losses (≤ 10 mm outer diameter (O.D.)) without vessel continuity. BMB identification and counting has a wide reported interrater reliability.[27] SWI scans were evaluated for BMB by four trained readers all blinded to clinical status using an identical protocol. The readers have had over 4 years experience evaluating SWI images for BMB using this protocol. Differences in BMB counts were resolved by a senior neuroradiologist (DK). Signal voids mimicking BMB, e.g. subarachnoid pial sulcal vessels, end-on vascular voids, rare angiomatous malformations as well as symmetrical focal basal ganglia iron deposits were excluded.

2.6 Statistical Methods

All data were entered into a relational database application (Access) and cross-verified. Numeric data was checked for normality. For comparing mean phase values, independent t-tests were used for between-group comparisons and paired t-tests were used to detect mean hemispheric differences and changes between first and last measures. The unadjusted risk ratios (RR) were determined from a contingency table. A 95% confidence interval (CI) for risk of dementia associated with BMB is presented. Logistic Regression was used to derive age adjusted Odds Ratios (OR) for progression to dementia associated with phase values. Age adjusted group-by-time changes in SWI phase values associated with cognitive status, were evaluated using the linear mixed model (SAS 9.2). All results assume an alpha = 0.05. All reported p-values are 2-tailed.

3.1 Enrollment and grouping

Of the 1,348 individuals screened, 112 participants were enrolled and initially classified into two groups according to cognitive status as normal (n=39) and MCI (n=73). Cognitive endpoints were defined as Normal, MCI, or Dementia and were assigned for each participant at the time of their last imaging study. Participants were classified into three groups based on their baseline and endpoint cognitive status and whether they stayed in the study long enough to obtain at least two SWI studies which were needed for serial iron analysis.

Group 1 (Normals; n=33) included participants who were classified as “normal” both neurologically and cognitively at baseline and “normal” at the time of the last MRI evaluation. Of the 39 initially enrolled, 37 remained at normal cognitive status for the entire study and 2 progressed to MCI and/or dementia and were therefore included in the MCI and Dementia groups. Four Normals left the study after only 1 MRI due to cancer (1) or the study ended before another SWI could be obtained (3). Note that ten of the participants classified into the Normal group were initially evaluated as MCI, however, they were reclassified to the Normal group since they were scored as normal in at least 3 subsequent cognitive evaluations (minimum of 3 and maximum of 7) over a range of 18 to 52 months.

Group 2 (MCI; n=23) included participants who were classified as MCI (n=73) or Normal (n=1) at baseline and remained at MCI status for the last MRI evaluation. Of the 73 initially enrolled, 46 remained at MCI status for the entire duration of the study while 27 progressed to dementia and were included in the Dementia group. Twenty-four of the remaining 46 MCI left the study or were excluded after only 1 imaging study due to co-morbidity (2), claustrophobia (1), loss of care support, transportation or moved (6), pacemaker insertion (1), no longer wanted to participate (3) or the study ended before another SWI could be obtained (11). One MCI died with autopsy-proven progressive supranuclear palsy and SWI findings reported.[10] The high morbidity and dropout rate of the MCI subjects has been noted by others and is a problem inherent to the clinical syndrome.[28, 29]

Group 3 (Dementia; n=27) included participants who were classified as MCI (n=26) or Normal (n=1) at baseline and progressed to dementia when evaluated for the last MRI study. The MCI who progressed to dementia did so at a progression rate of 0.15 per person-year of follow-up (95% CI: 0.09 to 0.21). Two participants from the Dementia group left the study or were excluded after only 1 useable imaging study due to loss of care support/moved (1) or no-longer willing to participate (1). The clinical picture of dementia in all of the cases was diagnosed as typical “Alzheimer’s disease” by experienced gerontologists and neurologists. The cognitive outcomes of our community-based MCI and control elderly cohorts over the 5.0 years (mean = 2.7 y) of the study conform with previous reports.[23]

3.1 Group description

Demographic data and neuropsychological profiles on the three groups whose cohorts are described in the above paragraphs are given in Table 1. No significant group differences were seen with respect to gender or years of education which may have influence on cognitive evaluations. The age at enrollment was significantly different between groups (p<0.001). Post hoc tests showed that the mean age of Normals was significantly lower than the Dementia group (p<0.001), but not the MCI group (p=0.056) and the MCI and Dementia groups were not significantly different (p= 0.14). The median number of cognitive evaluations and imaging studies obtained for each participant were similar for all three groups (Table 2). Participants received more cognitive evaluations during the study than MRI evaluations due to cost and scheduling issues. Normal participants stayed in the study longer (median of 35 months) compared to MCI (21 months) and demented (26 months) participants (Table 2). The time between MRI studies ranged from a minimum of 6 months up to 35 months, however, the median was similar for all three groups (Table 2). Only subjects with two or more scans have results reported and the times between scans are given in Table 2. The median interval in months for the “normals” was 14, MCI subjects 13, and demented 12.5.

3.2 Changes in regional iron levels over time

Of the 14 different regions evaluated, several regions in the right (WM, PG, CN, T, SN-pars reticularis) and left (P GP, RN-vascular) cerebral hemispheres showed a pattern of a subtle but not statistically significant decrease in SWI phase values in the Dementia cohort compared to MCI or Normals. These findings agree with previous reports of hemispheric asymmetries in iron content with the left hemisphere showing increased levels.[21] We have found, however, that after evaluation of the iron content of all ROIs only the left putamen showed a significant difference in mean SWI phase values between groups at baseline (ANOVA; p=0.043). A linear mixed model was then used to compare group-by time effects on phase values in the left putamen Figure 2 shows a significant left putaminal decrease of SWI phase values over the 50 month study duration in the 27 participants who progressed to dementia compared to Normals (p=0.035) and MCI (p=0.01). Kinetics of iron uptake are based on the change in phase units over time with a decrease in SWI phase units representing an increase in iron levels.[7] The stable MCI cohort had no change in putaminal iron over the course of the study compared to a small increase in iron for the Normals and a greater increase in the Dementia cohort.

3.3 Probability of dementia based on putaminal iron levels

We used logistic regression models to determine if left putaminal SWI phase data related to iron levels was predictive for progression to MCI status and found that age produced a higher and significant odds ratio (Odds ratio (OR) =1.16; p=0.002) than baseline phase values (OR=0.991; p=0.093) when used in the same model. Age was included since the mean age between Normals (74.9 ± 7.8; n=33) and all MCI (80.8 ± 4.9; n=49) at baseline was significantly different (p<0.001). When baseline left putaminal phase values were used separately as an independent variable for the model, an odds ratio close to one was determined (OR= 0.989; p=0.02). The results of these analyses show that age is a stronger predictor of dementia than putaminal phase values.

3.4 Number and regional distribution of BMB

Of the 33 normal participants, none were found to have BMB during the course of the study. Of the 23 in the MCI group, SWI detected one participant with greater than 1 BMB at entry which was still detectable 51 months later at the time of the last MRI. In the group that progressed to dementia, 8 of 26 had greater than 1 BMB at entry (range 2–69; median, 22) and an additional participant was found to have 5 at the time of their second SWI study obtained 8 months later. No BMB were detected in the remaining 15 in the demented group during the course of the study. The number of BMB increased in all 9 participants during the course of the study to a much greater extent in some than others (range of increase: 1–141; median, 16). Two participants showed a reduction of BMB at their last scan suggesting resorption of BMB during the course of the study. Among those classified as MCI at baseline, the Relative Risk (RR) for development of dementia associated with BMB >1 at any time is 2.06 (95% CI, 1.37–3.12).

As shown in Figure 3, significantly more BMB were detected with SWI in the right cerebral hemisphere than the left (p-value = 0.02, Wilcoxon Signed Ranks Test). The BMB are clustered primarily subcortically in the posterior parietal (23%), temporal (13%), frontal (15%), and occipital lobes (38%). The following BMB size distributions were found: 1–3 mm O.D. (95%), 4–5 mm O.D. (3%) > 5 but <10 mm O.D. (2%). An illustration of BMB as seen on SWI from one of the participants in the Dementia group is shown in Figure 4A and is compared to a T2* GRE image (Figure 4B).

3.5 Probability of dementia based on presence of BMB

Logistic regression was also used to determine whether the presence of BMB can predict whether the participants in our study in the MCI group will progress to dementia. When the presence or absence of BMB was included as an independent variable in the model to compare MCI (n=23) to Dementia (n=26), we found a significant odds ratio of 11.6 (p=0.026), however, the presence of BMB is significant only when the logistic regression model is not adjusted for age. When age and presence of BMB are included in the model, the odds ratio for neither variable was significant (OR=1.16 and 7.7, respectively; p=0.07 for each). The results of these analyses show an increased risk for progression to dementia when BMB are detected on SWI, however, a larger number of participants will need to be included to prove significance when adjusted for age due to the strong association of increased age and dementia.