Magnetization Transfer Ratio in Alzheimer Disease: Comparison with Volumetric Measurements

Subjects

Eighteen patients with AD and 18 healthy control subjects were recruited into the study. Diagnosis of probable AD was made according to the criteria of the National Institute of Neurologic and Communicative Disorders and Stroke and Alzheimer Disease and Related Disorders Association.¹⁷ Healthy control subjects had no subjective complaints of memory or other cognitive impairment and had no major neurologic, psychiatric, or systemic illness that might affect cognitive function. The local ethics committee approved the study. Informed consent was obtained from all participants and with the assent of next of kin of all patients with AD. The Mini-Mental State Examination (MMSE)¹⁸ was performed on all subjects. Based on the MMSE score, 12 patients with AD had mild disease severity (MMSE score range, 19–25), 4 had moderate (MMSE score range, 10–18), and 2 had severe (MMSE score range, 2–9). All control subjects scored above 27 of 30 on the MMSE.

MR Acquisition

MR imaging was performed on a 1.5T Signa scanner running software version 5.8 (GE Medical Systems, Waukesha, Wis). The manufacturer's quadrature head coil was used for both transmission and reception. Volumetric T1-weighted coronal images were obtained using an inversion-recovery prepared fast-spoiled gradient-echo technique (matrix, 256 × 192; FOV, 24 × 18 cm [reconstructed as 256 × 256 over 24 × 24 cm], TR, 14 ms; TE, 5.4 ms; TI, 650 ms; NEX, 1; FA, 15°, yielding 124 contiguous 1.5-mm sections).

In addition, 3D volumetric coronal proton attenuation-weighted spoiled gradient-echo images were obtained, with and without MT weighting (matrix, 256 × 192; FOV, 24 × 18 cm [reconstructed as 256 × 256 over 24 × 24 cm]; TR, 22.6 ms; TE, 5.4 ms; NEX, 0.75/FA, 15°, yielding 124 contiguous 1.5-mm sections). MT weighting was obtained by using a radio-frequency prepulse to partially saturate the broad resonance of immobile macromolecular protons. The MT pulse was a Hamming apodized 3-lobe sinc pulse with a duration of 6.4 ms and an equivalent on-resonance flip angle of 350°, applied 2 KHz off-resonance. The entire sequence was within specific absorption rate limits. Total scanning time for MT imaging was 18 minutes and 26 seconds.¹⁹

MR Image Analysis

For volumetric analysis, scans were transferred to Sun workstations (Sun Microsystems, Mountain View, Calif) and analyzed using the MIDAS image analysis program for image viewing and processing.²⁰ Scans were presented to the observer in random order, and all analyses were performed by persons blinded to subject details. The WB region (cerebral hemispheres, deep gray matter, cerebellum, and brain stem) was first extracted from skull, scalp, and other soft tissue using a semiautomated iterative morphologic technique. The region was then checked and manually edited where necessary to obtain a WB volume.²⁰ All WB volume measurements were performed by a single experienced rater with intrarater variability of 1% (ratio of the absolute difference in measurement to the mean in each subject). To correct for differences in subject head size, total intracranial volumes (TIV) were calculated according to a previously described protocol.²¹ WB volumes were standardized to mean TIV of control subjects. The standardization was carried out by using the slope of the relationship between WB volume and TIV estimated from a linear regression model relating WB volume to TIV, with both variables on logarithmic scales.²² For Hc volume measurements, all scans were first registered to a standard brain template using a 6 df algorithm to reduce any variability in landmarks used in delineating the Hc.²³ Each Hc was manually traced using multiple views to include the cornu ammonis, gyrus dentatus, and subiculum as described previously.²⁴ All Hc volume measurements were performed by one experienced rater with intrarater variability of 3%. Total (right + left) Hc volumes were calculated and adjusted to mean TIV of control subjects.²⁵ The standardization was carried in a fashion similar to that of the WB volume adjustment described above.

MTR Analysis

The MT prepared images were coregistered with the proton attenuation-weighted images by using a modified automated image registration (AIR) technique.^26,27 MTR maps, expressed in percentage units (pu),²⁸ were calculated pixel-by-pixel with the use of customized software according to the equation MTR = ([M_o − M_s]/M_o) × 100, where M_s and M_o represent the signal intensities with and without the saturation pulse. The MTR maps were registered to the volumetric scans by registering the proton attenuation-weighted images to the volumetric images and then applying the same transformation to the MTR maps.

Mean MTR was measured over the WB and in the Hc, pons, and parietal white matter (PWM) as volumes of interest (VOIs). The pons was chosen as a control region, in the vicinity of Hc but not typically affected in AD.²⁹ Although AD is a predominantly a gray matter disease, a control region in the white matter of the parietal lobe was outlined as the latter is less susceptible to CSF contamination, particularly in the presence of atrophy.

WB mean MTR was calculated by copying the segmented WB region from the volumetric image to the co-registered MTR map. Right and left Hc VOIs were outlined using MIDAS on approximately 15 coronal sections, avoiding the inclusion of CSF as much as possible. However, some CSF contamination was inevitable (Fig 1A). Right and left PWM VOIs were drawn in the coronal plane on approximately 4 sections (Fig 1B). Total Hc and PWM mean MTRs were calculated by averaging left and right mean MTRs for the Hc and PWM regions, respectively. A VOI was drawn in the center of the pons on approximately 4 coronal sections (Fig 1A).

• Download figure
• Open in new tab
• Download powerpoint

Fig 1.

Outlining volumes-of-interest in the coronal plane of magnetization transfer ratio maps of a patient with Alzheimer disease: left and right hippocampal and central pontine regions (A) and left and right parietal white matter regions (B).

In addition, histogram analysis was performed for WB VOIs. Pixels with 0 pu value (representing areas, mainly in background air regions, where signal-to-noise ratio in the original images was too low to calculate MTR) were first removed, and the remaining pixels were plotted as histograms with bin width of 0.1 pu. The histograms were normalized to correct for differences in brain volume and smoothed with a moving average window of 0.9 pu. The peak height, peak location, 25th percentile, 50th percentile (median), and 75th percentile points were then calculated from the histograms.

Statistical Analysis

Demographic, MMSE, and imaging data of patients with AD and control subjects were compared using the 2-sample t test, except for sex, where Pearson χ² test was used. A multiple logistic regression model was used to relate case-control status jointly to volumetric and MTR measures. However, a marked difference between the P values for the Wald and likelihood ratio tests suggested that the logistic regression model should not be relied on. Instead, to assess the extent to which volumetric and MTR measures were independently useful in differentiating patients with AD from control subjects, linear regression models were used relating each of these variables in turn to case-control status while adjusting for the other factor. Robust standard errors were used in these models to allow for differential heterogeneity between control subjects and cases. The ability of adjusted volume, mean MTR, and their combination to distinguish patients with AD from control subjects was assessed through the construction of receiver operating characteristic curves and calculation of sensitivity at specificity of 90%. Pearson correlation coefficients (r) were used to investigate the associations between MMSE and the various MR measures among patients with AD.

Subjects

AD patient and control groups were not significantly different in terms of age and sex (Table 1). Patients had significantly lower mean MMSE than control subjects.

MR Imaging Results

Volumetric measurements revealed that patients with AD had significantly smaller adjusted WB and total Hc volumes than control subjects (Table 1). The reductions are equivalent to approximately 10% for whole brain and 27% for Hc. In addition, patients with AD had significantly lower mean MTR for WB and total Hc but not PWM or pons (Table 1). WB MTR histogram analysis revealed that patients had significantly reduced 25th percentile point, but not peak height, peak location, median, or 75th percentile points.

From linear regression models with robust standard errors, WB adjusted volume differed significantly between patients and control subjects after adjustment for WB mean MTR (P < .0001). Further, WB mean MTR differed significantly between patients and control subjects after adjustment for WB adjusted volume (P = .008), demonstrating that both of these measures were independently associated with disease. There were no significant associations between WB adjusted volume and mean MTR among patients (r = 0.06, P = .81) or control subjects (r = 0.27, P = .27). Mean 25th percentile point did not differ significantly between patients and control subjects after adjustment for WB-adjusted volume (P = .22).

For Hc regions, the adjusted volume differed significantly between patients and control subjects after adjustment for Hc mean MTR using a linear regression model (P < .0001). However, Hc mean MTR did not significantly differ between patients with AD and control subjects after adjustment for total Hc-adjusted volume (P = .07). There was a significant correlation between total Hc-adjusted volume and Hc mean MTR among control subjects (r = 0.52, P = .03) but not among patients with AD (r = 0.33, P = .18).

Sensitivity and Specificity

Figure 2A demonstrates the superiority of measuring a combination of WB adjusted volume and mean MTR over either of these measures in isolation. With specificity set at 90%, the sensitivities of WB-adjusted volume and mean MTR in differentiating patients from control subjects were 50% and 33% respectively. The linear combination that best discriminates between patient and control groups was 10.8× adjusted brain volume (L) + WB mean MTR (pu). For this combination, the sensitivity was 83%. For the Hc region, measuring total volume and mean MTR had sensitivities of 89% and 61%, respectively, in differentiating patients and control subjects, with specificity set at 90% (Fig 2B). Measuring both yielded a sensitivity of 89%, similar to that of total Hc volume alone.

• Download figure
• Open in new tab
• Download powerpoint

Fig 2.

Receiver operating curves plotting the sensitivity and specificity of differentiating Alzheimer disease patients from control subjects by measuring adjusted volume, mean magnetization transfer ratio or both for whole brain (A) and hippocampal regions (B).

Associations between MR Measures and MMSE

Table 2 shows the correlation coefficients for the relationships between MMSE and various volumetric and MTR measures among patients with AD. Only adjusted WB volume was significantly associated with MMSE (r = 0.47, P = .048) (Fig 3).

• Download figure
• Open in new tab
• Download powerpoint

Fig 3.

Relationship between adjusted whole brain volume and MMSE score among patients with AD and control subjects. The line indicates the linear regression of adjusted whole brain volume on MMSE score among patients with AD (correlation coefficient = 0.47, P = .048).