Subjects
Forty-one patients who underwent MRI with diagnosis of AD or mild cognitive impairment (MCI) due to AD between September 2015 and March 2016 were recruited. Patients with AD were included if they met the criteria for probable AD established by the National Institute of Neurological and Communicative Disorders and Stroke-Alzheimer Disease and Related Disorders Association (NINCDS/ADRDA) [21]. Patients were excluded from the study if they had a significant history of psychiatric or neurological disorders other than AD, including stroke, head injury, epilepsy, psychiatric disorders, alcohol abuse, or other serious medical conditions. All patients underwent MR scanning at 3.0 T and standard dementia screening, which included a medical history check, Mini-Mental State Examination (MMSE), neuropsychological testing, and MR imaging. Cognition was assessed using the MMSE, which evaluates general cognitive function, including orientation to time and place, attention and calculation, language, and memory [22]. Among these, 28 patients also underwent brain perfusion SPECT using N-isopropyl-4-[123I]iodoamphetamine (123I–IMP) within 1 month of MR scanning at our hospital. This tracer has been reported to be feasible for the quantitative evaluation of CBF in routine clinical practice [23]. Most of others underwent SPECT using other tracer or at another hospital, but we did not include their SPECT images for the analysis in this study. We used all available images of both ASL and SPECT obtained from all patients.
MR imaging
Images were acquired using a 3.0 T MR system (Discovery750w, GE Medical Systems) and a 12-channel head coil.
Anatomic information was obtained from a sagittal three-dimensional (3D) T1-weighted sequence, (the parameters include: TR = 6.6 ms, TE = 2 ms, 14° flip angle, matrix = 256 × 256, 170 stions, voxel size = 1.0 × 0.9 × 0.9 mm3, FOV = 23 × 23 cm), acquisition time was 6 min 0 sec.
The ASL sequence consisted of a 3D, multi-delay PCASL, with a fast spin-echo acquisition with background suppression. The labeling plane was set at the base of the brain without the information of MR angiography. The imaging protocol of PLD1.5 was as follows: TR = 4641 ms, TE = 10.7 ms, locations = 36, FOV = 23 × 23 cm, voxel size = 2 × 2 × 4 mm3, PLD = 1.5 s, labeling duration = 1.5 s, number of excitations (NEX) = 1, acquisition time was 1 min 33 s. The imaging protocol of PLD2.5 was as follows: TR = 5336 ms, TE = 10.7 ms, PLD = 2.5 s, NEX = 2, all other parameters maintained the same, and the acquisition time was 2 min 51 s.
A two-compartment model with finite labeling duration was used for PCASL quantification. An approximately proton-density-weighted image was obtained by turning the labeling RF off. Calculation of flow was based on the following equation.
$$ \mathrm{f}=\frac{\lambda \left({S}_{ctrl}-{S}_{lbl}\right)\left(1-{e}^{-\frac{t_{sat}}{T_{1 g}}}\right)}{2\alpha {T}_{1 b}\left(1-{e}^{-\frac{\tau}{T_{1 b}}}\right){S}_{ref}}{e}^{\frac{w}{T_{1 b}}} $$
(1)
where f is the measured CBF; S is the signal from the control, label, or reference image as determined by the subscript, S
lbl is the label image, i.e. image obtained with unbalanced RF labeling that gives rise to perfusion weighting, S
ctrl is the control image, i.e. image obtained with balanced RF so that the arterial blood is not labeled, S
ref is the proton density image that is obtained with labeling RF turned off; T
1
b is the T1 of blood; T1
g is the T1 of gray matter; α is the labeling efficiency; λ is the brain–blood partition coefficient; t
sat is the saturation time for proton density images (2 s); τ is the labeling duration (1.5 s); and ω is the post label delay. We used a gray matter T1 estimate of 1.2 s and an assumed blood T1 of 1.6 s [24]. The labeling efficiency was assumed to be 0.8 for the PCASL.
Preprocessing and MR imaging data analysis
Data analyses were carried out by using Statistical parametric mapping 12 (SPM12) (http://www.fil.ion.ucl.ac.uk/spm/software/spm12/). Both 3D T1-weighted and 3D PCASL images were corrected for image distortion due to gradient non-linearity using ‘GradWarp’ [25]. Preprocessing of 3D T1-weighted images consisted of realignment, coregistration, and segmentation.
ASL images were linearly registered to the brain extracted from the 3D T1-weighted images. Mean whole-brain CBF values were calculated in the brain mask, converted to quantitative CBF maps in the unit of mL/100 g/min, spatially normalized to the Montreal Neurological Institute (MNI) space with a 2-mm isotropic resolution, and smoothed with an isotropic kernel of 6 mm. Complementary voxel-wise comparisons of 2 kinds of CBF maps were performed by SPM12 as well. Correlation results were statistically thresholded at p < 0.001 uncorrected. All results were shown in the MNI space. The locations with significant results were determined using the “Neuromorphometrics” function of SPM12.
Brain perfusion SPECT
SPECT was performed by intravenous injection of 148 MBq 123I–IMP (Nihon Mediphysics, Hyogo, Japan) in participants seated at rest with their eyes open. A dual-head gamma camera with integrated thin-slice diagnostic CT (Symbia® T16, Siemens Healthcare, Molecular Imaging, Hoffman Estates, IL, USA) was used. The SPECT scans were acquired using low-medium-energy general-purpose collimation, a 128 × 128 matrix of 3.3-mm pixel size, and a total of 300 s/rotation in a continuous-rotation mode. Subsequent to the SPECT acquisition, a reduced-dose CT scan was acquired with 130 kV and 150 ref. mAs. The CT data were generated with a 3-mm slice thickness using a smooth reconstruction kernel (H08s, Siemens Healthcare) and a 2-mm slice thickness using a medium kernel (H31s medium sharp, Siemens Healthcare).
SPECT reconstruction was performed using filtered back projection using a Butterworth filter with cutoff = 0.35/cm and order 8. A uniform attenuation correction was performed using Chang’s method with μ = 0.11.
Statistical analysis
Voxel-based analyses were performed for PLD1.5, PLD2.5 and brain perfusion SPECT using MMSE scores as covariates. Differences of CBF between PLD1.5 and PLD2.5 were assessed by paired t-test. The individual brain images were normalized by the global values. The p value threshold was 0.001 at the voxel level, and the regions with the extent under the expected voxels per cluster were omitted. The plots of the correlation between the voxel values and MMSE scores at the most significant area are provided. These were performed by SPM12.