Combination of MRI hippocampal volumetry and arterial spin labeling MR perfusion at 3-Tesla improves the efficacy in discriminating Alzheimer's disease from cognitively normal elderly

Unc

orre

cted

Aut

hor P

roof

Journal of Alzheimer’s Disease xx (20xx) x–xxDOI 10.3233/JAD-131868IOS Press

1

Combination of MRI HippocampalVolumetry and Arterial Spin Labeling MRPerfusion at 3-Tesla Improves the Efficacyin Discriminating Alzheimer’s Diseasefrom Cognitively Normal Elderly Adults

1

2

3

4

5

Henry Ka-Fung Maka,f,∗, Wenshu Qiana, Kwok Sing Ngb, Queenie Chanc, You-Qiang Songd,f ,Leung Wing Chue,f and Kelvin Kai-Wing Yaug

6

7

aDepartment of Diagnostic Radiology, The University of Hong Kong, Hong Kong SAR, China8

bLi Ka Shing Faculty of Medicine, The University of Hong Kong, Hong Kong SAR, China9

cPhilips Healthcare, Hong Kong SAR, China10

dDepartment of Biochemistry, The University of Hong Kong, Hong Kong SAR, China11

eDepartment of Medicine, The University of Hong Kong, Hong Kong SAR, China12

f Alzheimer’s Disease Research Network, The University of Hong Kong, Hong Kong SAR, China13

gDepartment of Management Sciences, City University of Hong Kong, Hong Kong SAR, China14

Handling Associate Editor: J. Wesson Ashford

15

Accepted 18 February 2014

Abstract.16

Background: Structural magnetic resonance imaging has been employed for evaluation of medial temporal atrophy in patientswith Alzheimer’s disease (AD). Arterial spin labeling (ASL) technique could detect cerebral perfusion abnormalities in AD.

17

18

Objective: We hypothesized that combination of hippocampal volumetry and cerebral blood flow yield higher accuracy thaneither method alone in discriminating AD patients from cognitively normal older adults.

19

20

Materials and Methods: 13 AD patients and 15 healthy controls were studied using a 3-tesla scanner. Standardized T1W 3Dvolumetric Fast Field Echo and QUASAR ASL sequences were employed for cerebral volumetry and perfusion respectively.Manual right and left hippocampal volumetry was performed by ANALYZE software, with total intracranial volume normaliza-tion. ASL data were analyzed by institutional specially-design software to calculate cerebral blood flow of region-of-interestsplaced at the middle and posterior cingulate gyri.

21

22

23

24

25

Results: Right and left hippocampal volumes and middle and posterior cingulate gyri cerebral blood flows were significantlylower in the patients than in the controls (independent-samples t-tests, p < 0.05), and prediction accuracies of 89.3%, 82.1%,75.0% and 71.4% were achieved for each of the above parameters, respectively. In distinguishing patients from controls usingcorresponding optimized cut-off values, various combinations of these parameters were used to create the Receiver OperatingCharacteristic curves. The highest area under curve value was 0.944, by combining cerebral blood flow at the middle cingulategyrus, normalized right and left hippocampal volumes.

26

27

28

29

30

31

Conclusions: A ‘one-stop-shop’ magnetic resonance study of combined hippocampal volumetry and cerebral perfusion hasimproved efficacy in discriminating AD patients from cognitively normal older adults.

32

33

Keywords: Alzheimer’s disease, arterial spin labeling, cerebral blood flow, cerebral perfusion, hippocampal volumetry, magneticresonance imaging, medial temporal atrophy

34

35

∗Correspondence to: Henry Ka-Fung Mak, Department of Diag-nostic Radiology, The University of Hong Kong, Room 406, BlockK, Queen Mary Hospital, 102 Pokfulam Road, Hong Kong. Tel.:+852 28170373; Fax: +852 28174013; E-mail: [email protected].

ISSN 1387-2877/14/$27.50 © 2014 – IOS Press and the authors. All rights reserved

Unc

orre

cted

Aut

hor P

roof

2 H.K.-F. Mak et al. / MRI Volumetry & ASL in AD and Normal Controls

INTRODUCTION36

In the past decade, structural magnetic resonance37

imaging (MRI) has been increasingly employed to38

aid the diagnosis of Alzheimer’s disease (AD) and39

other forms of dementia [1]. Numerous cross-sectional40

studies have found that AD patients have a reduction41

in the volumes of medial temporal structures, such42

as the entorhinal cortex and the hippocampus when43

compared with controls [2]. The supportive role of44

medial temporal lobe atrophy in the revised criteria of45

National Institute of Neurological and Communicative46

Disorders and Stroke and the Alzheimer’s disease and47

Related Disorders Association (NINCDS-ADRDA)48

for the diagnosis of AD had been proposed [3].49

A recent study [4] compared the efficacy of two reg-50

istration methods of computer-assisted hippocampal51

volumetry with manual hippocampal volumetry in a52

Chinese cohort of AD patients and cognitively nor-53

mal older adults (CN). The study showed that both54

registration methods have similar efficacies as manual55

volumetry in distinguishing AD from CN and achieved56

overall high prediction accuracies of 87.2% and 84.6%57

for manual right hippocampal (RH) and left hip-58

pocampal (LH) volumetry. In particular, a significant59

reduction (20–30%, p < 0.001) in all normalized hip-60

pocampal measurements when comparing AD patients61

to normal elderly subjects was found.62

Quantitative MRI studies on hippocampal size in63

mild cognitive impairment (MCI) using both region-64

of-interest and voxel-based volumetric studies have65

been less successful. The overall correct classification66

of normal and MCI subjects attained with hippocampal67

and entorhinal volumes varied from 60 to 85% [5].68

Radionuclide studies using [18F] 2-69

fluorodeoxyglucose positron emission tomography70

(18F-FDG PET) and single photon emission computed71

tomography (SPECT) have been used respectively72

to study altered cerebral glucose metabolism and73

cerebral perfusion in AD, with complementary results74

indicating coupling of blood flow and metabolism in75

the brain. However, PET/SPECT studies are inherently76

invasive, as they involve the intravenous injection of77

radioactive tracers.78

Arterial spin labelling (ASL) is a quantitative MRI79

perfusion imaging method which has the advantage80

of being non-invasive, as it does not require intra-81

venous injection of any exogenous tracer [6]. ASL MRI82

uses electromagnetic labeling of the naturally occur-83

ring water in blood to acquire images sensitive to flow.84

A preliminary study by Alsop et al. [7] evaluated the85

utility of ASL MRI for detection of cerebral blood86

flow (CBF) abnormalities in AD. The study found that 87

cognitive decline correlates with a decrease in CBF 88

in the posterior parietal and posterior cingulate (PC), 89

but not in temporal cortices. Another study showed 90

regional hypoperfusion in brain regions of AD patients 91

similar to those seen in 18F-FDG PET and HMPAO 92

SPECT studies of similar populations [8]. The same 93

research group found different patterns of hypoperfu- 94

sion in distinct subgroups of MCI, further suggesting 95

that cognitive impairment is related to hypoperfusion 96

[9]. 97

The pulsatile ASL method with multiple-delay time 98

sampling, i.e., quantitative STAR labeling of arte- 99

rial regions (QUASAR) pulse sequence [6] had been 100

employed in a previous study [10] for detection of 101

patients with AD, with 74% sensitivity and 80% speci- 102

ficity based on absolute CBF in the PC gyri and 103

precunei (threshold- 26.0 ml/100 g/min). Using similar 104

methodology, a previous study by our group demon- 105

strated that significant reduction in CBF (20–28%, 106

p < 0.05) in the middle cingulate (MC) and PC in a 107

cohort of moderate AD, as compared to CN [11]. 108

Advanced neuroimaging methods have been advo- 109

cated as biomarkers in differentiating MCI, mild AD, 110

or AD patients from CN. A large cohort of 80 healthy 111

control subjects and 68 AD were recruited from the 112

Alzheimer’s Disease Neuroimaging Initiative (ADNI) 113

[12] to study the relative capability of MR imaging 114

and FDG PET to depict changes for discriminating 115

the two groups. The Receiver Operator Characteris- 116

tic (ROC) analysis of MR imaging (hippocampus) and 117

FDG PET (entorhinal cortex) showed Area-Under- 118

Curve (AUC) values of 0.899 and 0.706, respectively. 119

Multimodality imaging using MRI and FDG-PET 120

had been employed by Jhoo et al. [13] in discrim- 121

ination of normal aging, MCI, and AD. The model 122

combining all three parameters (hippocampal volume, 123

parahippocampal cingulum fractional anisotropy, and 124

glucose metabolism) yielded a discrimination accu- 125

racy of 94.1% for normal controls versus AD. Walhovd 126

et al. [14] combined FDG-PET, MR morphometry, and 127

diffusion tensor imaging derived fractional anisotropy 128

measures to aid diagnosis of MCI, and achieved diag- 129

nostic accuracy ranging from 65.6% (parahippocampal 130

gyrus) to 73.4% (inferior parietal cortex). 131

In current study, we sought to evaluate the efficacy 132

of manual hippocampal volumetry, ASL-MRI derived 133

CBF, and their combination in discriminating a cohort 134

of AD from CN. Our hypothesis is that the combi- 135

nation of structural MRI hippocampal volumetry and 136

ASL MRI perfusion might achieve higher efficacy in 137

diagnosis than using individual method.

Unc

orre

cted

Aut

hor P

roof

H.K.-F. Mak et al. / MRI Volumetry & ASL in AD and Normal Controls 3

Table 1The clinical, genetic, and psychological profile of the AD subjects and normal controls

No. Age Gender Ed Level APOE4 (1 = −ve; 2 = +ve) APOE alleles (2, 3, or 4) MMSE ADAS-cog

AD subjects 1 56 0 0 1 3,3 16 43.002 70 1 8 1 3,3 23 14.673 73 0 5 2 3,4 10 38.674 73 0 4 2 3,4 16 32.335 75 0 0 1 3,3 21 12.676 77 0 0 1 3,3 11 33.007 78 1 16 2 3,4 17 22.348 78 0 0 1 3,3 15 23.339 79 1 9 2 3,4 12 27.0010 79 0 4 2 3,4 11 18.6611 79 0 4 1 3,3 22 17.6612 81 0 0 1 3,3 16 Nil13 82 0 8 2 2,4 22 19.67

Controls 1 60 0 4 1 3,3 30 Nil2 62 0 11 1 3,3 30 1.003 64 0 1 1 2.3 30 Nil4 67 1 9 1 2.3 25 3.675 69 0 3 1 3,3 27 9.676 69 0 12 1 3,3 27 7.337 71 0 3 1 3,3 30 3.008 71 0 12 2 2,4 29 5.009 72 0 0 1 2,3 24 8.0010 72 0 0 1 3,3 29 5.3311 73 0 1 1 3,3 27 7.3312 75 0 1 2 3,4 30 5.0013 76 0 3 1 3,3 29 7.3314 80 0 3 1 2.3 30 7.3315 81 0 9 1 3.3 30 1.66

p 0.068 0.251 0.849 0.052 0.036 <0.001 <0.001

MATERIALS AND METHODS138

Subjects139

Participants in this retrospective study consisted of140

20 CN (17 females, 3 males) and 20 AD patients (15141

females, 5 males) who met the criteria of AD. Based on142

these subject groups, a separate study was published on143

using voxel-based morphometry as imaging biomarker144

in AD [4].145

This was a nested case-control sample of an ongoing146

local cohort study on aging and dementia, and details147

on clinical and neuropsychological assessments of par-148

ticipants were published previously [15]. A detailed149

medical history and Mini-Mental State Examination150

(MMSE) and Alzheimer’s Disease Assessment Scale151

cognitive subscale scores (ADAS-cog) for global cog-152

nitive ability, and blood test for apolipoprotein E4153

(APOE4) genotype were obtained from all partici-154

pants. A diagnosis of probable AD was made using155

criteria from NINCDS-ADRDA [16]. Informed con-156

sent was obtained from all subjects who participated157

in the study and Institutional Review Board approval158

was obtained prior to commencement of the study.159

The final cohort comprised 15 CN (14 females, 1 160

male) and 13 AD patients (10 females, 3 males). One 161

AD patient was excluded based on final diagnosis of 162

cerebral amyloid angiopathy (see MRI scanning pro- 163

tocol), and 6 AD patients and 5 CN were excluded 164

due to poor ASL image quality as a result of motion 165

and distortion by susceptibility artifacts (as detailed 166

in [11]). Statistical analysis of the final cohort showed 167

no significant differences in age, gender, and education 168

level (number of years) between AD and CN (Table 1), 169

but a significant difference in the APOE4 allele status 170

(p = 0.036). 171

MRI acquisition 172

All MRI examinations were performed using a 173

3.0-tesla scanner (Achieva, Philips Healthcare, the 174

Netherlands), with an 8-channel head coil. For MRI 175

volumetric studies, a standardized T1W 3D vol- 176

umetric Fast Field Echo sequence was employed 177

(TR/TE 7.0/3.2ms, TI 869 ms, flip angle 8◦, voxel size 178

1 × 1 × 1 mm3, FOV 256 mm, 156 contiguous slices). 179

Axial T2-weighted fast spin-echo images (TR/TE 180

3000/80 ms, flip angle 90◦, slice thickness 5 mm, ETL 181

Unc

orre

cted

Aut

hor P

roof

4 H.K.-F. Mak et al. / MRI Volumetry & ASL in AD and Normal Controls

16) were obtained to exclude structural abnormalities.182

Almost all AD patients and all CN showed no sig-183

nificant signal abnormalities on T2W images. Only184

non-specific punctate T2 hyperintense foci were found185

in basal ganglia, subcortical and deep white matter of186

some of them. One AD subject was excluded since187

there were diffuse T2 hypointense signal abnormalities188

involving bilateral cerebral hemispheres, with spar-189

ing of brainstem and cerebellar hemispheres. The final190

diagnosis of cerebral amyloid angiopathy was made191

and presence of diffuse microbleeds confirmed by sus-192

ceptibility imaging.193

The QUASAR [6] sequence consisted of a Look-194

Locker-based Echoplanar Imaging sequence with195

the following parameters: TR/TE 4000/22 ms, FOV196

240 mm, matrix 64 × 64, � = 35◦, 7 axial 6 mm slices197

with 2 mm gap, labeling slab 150 mm, with a gap to the198

closest slice of 15 mm. The first image was acquired199

at a delay of 40 ms, while 13 consecutive time points200

were measured at intervals of 300 ms, thus yielding ten201

time points ranging from 40 up to 3640 ms. The dura-202

tion of the ASL acquisition was approximately 6 min203

[17].204

Data analysis205

Details on the data analysis can be found in our pub-206

lished studies [4, 11]. A brief summary of the data207

processing was described below.208

Manual hippocampal volumetry209

T1-weighted coronal images reconstructed as210

1.5 mm slices perpendicular to the long axis of the211

hippocampus were measured. The boundaries of both212

hippocampi were traced manually by an experienced213

neuroradiologist (HKM) and their volumes determined214

by ANALYZE 7.0 software (Mayo Clinic Founda-215

tion, Rochester, MN, USA). Normalization with total216

intracranial volume was also performed.217

ASL218

The ASL data were analyzed using institutional219

specially-design software [6, 17] to calculate the220

CBF, and longitudinal relaxation time (R1) maps.221

The preprocessing and statistical analyses were car-222

ried out with SPM8 (statistical parametric mapping223

software, Wellcome Trust Centre for Neuroimaging,224

London, UK) implemented in MATLAB (Mathworks225

Inc., Natick, MA, USA). Preprocessing included co-226

registration and normalization of perfusion parametric227

maps to the standard space of the Montreal Neurolog-228

ical Institute (MNI) brain. R1 map was co-registered 229

to the T1W images, followed by spatial normalization. 230

The normalization parameters derived were applied to 231

the CBF maps of the same subject. The CBF maps 232

were then spatially smoothed with an isotropic 8 mm 233

full width at half maximum Gaussian filter. 234

Based on the Regions-of-Interest (ROI) results of 235

our published study [11], statistical significant reduc- 236

tion of mean CBF in the MC and PC gyri were found. 237

Hence, only these regions were chosen in current study. 238

Statistical analysis 239

The mean values and standard deviations of CBF 240

in MC and PC gyri and normalized hippocampal vol- 241

umes of both AD and CN groups were calculated. 242

Independent-samples t-test was used to test for any 243

significant differences of these parameters between the 244

two groups. The level of significance was p < 0.05. 245

The sensitivity, specificity, and accuracy of the com- 246

bined modality model versus single modality model 247

were computed using commercially-available software 248

(SPSS version 19.0.0, Chicago, IL, USA). 249

Various combinations of these neuroimaging 250

biomarkers were used to create the ROC curves, 251

with AUC measured. Youden index (= Sensitivity + 252

Specificity-1) was employed to select the optimal cut- 253

offs of various parameters. 254

Pearson correlations and multiple linear regression 255

analyses of the clinically-observed severity measure- 256

ments (MMSE, ADAS-cog, and severity index) with 257

brain measurements were computed: Severity index 258

(SI) = 0.5*(MMSE + 30 – 2 × ADAS-cog). 259

Scatter-plot of MC and PC CBF, and normalized RH 260

and LH volumes against the SI were shown (Fig. 4). 261

RESULTS 262

The demographic, genetic (APOE4), and neuropsy- 263

chological profiles of the AD and CN were tabulated 264

(Table 1). 265

The mean (standard deviation) in AD and CN were, 266

respectively: 102.9 (32) and 153.8 (21) for normal- 267

ized RH; 107.7 (32.4) and 149.0 (17.5) for normalized 268

LH; 42.0 (11.0) and 52.4 (11.3) for MC CBF; 35.0 269

(13.4) and 48.6 (16.9) for PC CBF. The p-values 270

of the independent-samples t-tests of these parame- 271

ters between patient group and controls were <0.001, 272

<0.001, 0.021 and 0.025, respectively (Table 2). 273

Various combinations of these parameters were used 274

to create the ROC curves and their AUC (Figs. 1–3).

Unc

orre

cted

Aut

hor P

roof

H.K.-F. Mak et al. / MRI Volumetry & ASL in AD and Normal Controls 5

Table 2The mean values and standard deviation of cerebral blood flow in middle (MC) and posterior (PC) cingulate gyri and normalized hippocampal

volumes on both sides of patients and controls

0: control (15) Range Mean Standard. Deviation Independent- samples T-test1: patient (13) p-value

MC 0 34.478∼78.032 52.427 11.312 0.021CBF* 1 21.982∼59.294 42.036 10.972PC 0 30.777∼76.957 48.572 16.873 0.025CBF 1 16.829∼60.987 34.973 13.402Normalized Vol RH 0 1.070∼1.820 1.538 0.210 <0.001

1 0.432∼1.427 1.029 0.320Normalized Vol LH 0 1.221∼1.753 1.490 0.175 <0.001

1 0.498∼1.427 1.077 0.324

*The unit of CBF is in ml/100 g/min.

Fig. 1. Single parameter: CBF of middle and posterior cingulate gyriand the normalized volume ratios (percentage*100) of right and lefthippocampi.

The highest AUC value was 0.944, by combining275

normalized RH, LH, and MC CBF; AUC value of indi-276

vidual parameter such as normalized RH, normalized277

LH, MC CBF, and PC CBF was 0.928, 0.872, 0.749278

and 0.744, respectively. (Figs. 1–3, Table 3)279

Using corresponding optimized cut-off values, pre-280

diction accuracies of 89.3%, 82.1%, 75.0% and 71.4%281

for RH volumetry, LH volumetry, MC CBF, and PC282

CBF, respectively, were achieved in distinguishing AD283

from CN. Combination of 2, 3 or 4 of these parame-284

ters achieved accuracies ranging from 82.1% to 89.3%285

(Table 3).286

In the entire cohort of subjects plus controls, there287

were statistically significant correlations (p < 0.05)288

of the clinically observed severity measurements289

Fig. 2. Combining two parameters. MR: M cingulate with rVol RH;ML: M cingulate with rVol LH; PR: P cingulate with rVol RH; PL:P cingulate with rVol LH.

(MMSE, ADAS-cog, and SI) with brain scan param- 290

eters (normalized RH and LH volumes, MC and PC 291

CBF). For example, Pearson correlation between SI 292

and the brain scan measurements ranged from 0.502 293

to 0.646 (Table 4). Such strong correlations were also 294

demonstrated across diagnostic groups (Fig. 4). 295

Applying multiple regression analysis to the entire 296

cohort, we showed that combination of brain scan 297

parameters such as MC CBF, RH and LH volumes 298

achieved higher significance in prediction of clinically 299

observed severity measurements (Table 4). 300

DISCUSSION 301

Our current study evaluated the diagnostic per- 302

formance of manual hippocampal volumetry versus 303

Unc

orre

cted

Aut

hor P

roof

6 H.K.-F. Mak et al. / MRI Volumetry & ASL in AD and Normal Controls

Fig. 3. Combining three or more parameters. MRL: M cingulatewith rVol RH and rVol LH; PRL: P cingulate with rVol RHand rVol LH; MPR: M&P cingulate with rVol RH; MPL: M&Pcingulate with rVol LH; MPRL: M&P cingulate with rVol RH andrVol LH.

pulsatile ASL derived CBF, and their combination in304

discriminating a cohort of AD from CN. Our findings305

confirmed the diagnostic roles of these MRI techniques306

and highlighted the emerging role of ASL CBF.307

First, our current study employed a smaller cohort 308

of 13 AD and 15 CN [11] than our published study 309

[4] to allow comparison between manual hippocampal 310

volumetry and ASL CBF methods. In a prior study of 311

a larger cohort of 19 AD and 20 CN, the prediction 312

accuracy and AUC were 87.2% and 0.94 and 84.6% 313

and 0.93 for normalized RH and LH volume, respec- 314

tively. Similar diagnostic efficacies were achieved in 315

current study, with prediction accuracy and AUC being 316

89.3% and 0.928 and 82.1% and 0.872 for normalized 317

RH and LH volume, respectively. These findings vali- 318

dated the robust discriminatory power of hippocampal 319

morphometry. 320

Second, pulsatile ASL MRI was found in our study 321

to be inferior to hippocampal volumetry in diagnos- 322

tic accuracy, with prediction accuracy and area under 323

ROC curve being 75% and 0.749 and 71.4% and 0.744 324

for MC and PC CBF. A previous study using QUASAR 325

ASL [10] showed significant differences of CBF in the 326

PC and precunei. Based on the mean CBF in these 327

ROIs, the study achieved 73.9% sensitivity, 80.0% 328

specificity, and 76.7% accuracy in discriminating 20 329

AD and 23 CN, similar to our respective results of 330

61.5%, 86.7% and 74.9% (based on CBF in the MC 331

gyrus). A recent study [18] employed ‘qualitative’ 332

evaluations of continuous ASL and high resolution 333

T1W images by four blinded experienced readers and 334

found that perfusion abnormalities were superior to 335

structural volume losses in early detection of AD, 336

Fig. 4. Scatter-plot of middle and posterior cingulate CBF, and normalized right and left hippocampal volumes against the severity index.

Unc

orre

cted

Aut

hor P

roof

H.K.-F. Mak et al. / MRI Volumetry & ASL in AD and Normal Controls 7

Table 3The sensitivity, specificity, and accuracy results of the single parameter versus combined parameter models, optimal cut-offs of various parameters

based on Youden index, and AUC of the Receiver Operating Characteristic (ROC) curves of various combinations of parameters

cut-off Sensitivity Specificity Youden index Accuracy AUCmean 95% CI

MC 44.301 0.615 0.867 0.482 0.750 0.749 0.565, 0.933PC 31.097 0.462 0.933 0.395 0.714 0.744 0.560, 0.927rVol RH 1.446 1 0.800 0.800 0.893 0.928 0.831, 1.000rVol LH 1.396 0.923 0.733 0.656 0.821 0.872 0.743, 1.000MR 0.322 (M = 45.184, R = 1.427) 1 0.800 0.800 0.893 0.938 0.854, 1.000ML 0.757 (M = 36.462, L = 1.331) 0.615 1 0.615 0.821 0.887 0.769, 1.000PR 0.348 (P = 31.219, R = 1.427) 1 0.800 0.800 0.893 0.933 0.841, 1.000PL 0.426 (P = 33.726, L = 1.404) 0.923 0.867 0.790 0.893 0.923 0.825, 1.000MRL 0.299 (M = 44.799, R = 1.464, L = 1.404) 1 0.800 0.800 0.893 0.944 0.863, 1.000PRL 0.439 (P = 31.026, R = 1.371, L = 1.435) 0.923 0.867 0.790 0.893 0.938 0.852, 1.000MPR 0.345 (M = 45.184, P = 33.755, R = 1.427) 1 0.800 0.800 0.893 0.933 0.845, 1.000MPL 0.449 (M = 44.799, P = 31.219, L = 1.404) 0.846 0.867 0.713 0.857 0.908 0.801, 1.000MPRL 0.316 (M = 45.184, P = 33.726, R = 1.464 L = 1.404) 1 0.800 0.800 0.893 0.938 0.854, 1.000

Table 4Pearson correlations and multiple linear regressions of the clinically-observed severity measurements with brain measurements in whole group

(AD subjects plus controls)

Pearson correlation Multiple linear regressionsMC PC rVol RH rVol LH MC + RH + LH

MMSE r 0.536 0.435 0.757 0.718 0.785p 0.003 0.021 <0.001 <0.001 <0.001

ADAS cog r −0.528 −0.499 −0.599 −0.567 0.659p 0.007 0.011 0.002 0.003 0.007

Severity index r 0.539 0.502 0.646 0.617 0.698p 0.005 0.011 <0.001 0.001 0.002

MC, middle cingulate CBF, RH, normalized right hippocampal volume, LH, normalized left hippocampal volume.

with higher sensitivity (85% versus 56%) and accuracy337

(70% versus 63%). The disparity was likely due to the338

poor sensitivity of visual assessment of high resolution339

structural images, as suggested by the study.340

Third, in the literature, the relative capability of341

CBF to detect AD (versus structural MRI) was mainly342

based on FDG-PET. However, conflicting results were343

obtained in these studies. De Santi et al. [19] found344

that regional measures of glucose metabolism had345

greater diagnostic accuracy than manual MRI volume-346

try. The diagnostic classification accuracies of PET347

ranged from 87% to 100%, while MRI regional volume348

ranged from 78% to 83%. Kawachi et al. [20] found349

that the diagnostic accuracy and AUC were 88.5% and350

0.953, and 83% and 0.910 for FDG PET (bilateral PC351

and right temporoparietal area) and VBM-MRI (bilat-352

eral amygdala/hippocampus complex), respectively.353

Nevertheless, our results were similar to a large-scale354

study [12] of ADNI using automated methods, which355

found that the areas under ROC were 0.899 (95% CI356

- 0.847, 0.951) and 0.706 (95% CI - 0.621, 0.792) for357

MRI hippocampus and FDG PET entorhinal cortex,358

respectively. Another large-scale study based on ADNI359

dataset showed that automated MRI structural mea- 360

sures achieved slightly higher classification accuracy 361

than FDG-PET measures, being 85% versus 82.5% 362

[21]. In a multimodal imaging study in MCI [14], it 363

was also recognized that combinations of FDG-PET 364

measures did not provide an independent contribu- 365

tion to diagnostic accuracy once MRI morphometric 366

measures were taken into account [12]. 367

Fourth, unlike prior investigators who determined 368

the cutoff values in a relatively subjective manner (such 369

as 2 standard deviation below the means of the con- 370

trol groups [18]), we adopted Youden index, which 371

jointly reflects sensitivity and specificity, as a scien- 372

tific measure to derive the optimal cutoffs for our local 373

cohort. Moreover, we found that combination of MRI 374

hippocampal volumetry and ASL CBF improved the 375

diagnostic efficacy, as compared to the single parame- 376

ter of normalized RH volume (AUC being 0.928). The 377

highest achievable AUC (0.944) was by a combination 378

of three parameters (normalized RH and LH volumes 379

and MC CBF) instead of all four. A combination of two 380

parameters (RH volume and MC CBF) also improved 381

the diagnostic yield, with AUC of 0.938, being the 382

Unc

orre

cted

Aut

hor P

roof

8 H.K.-F. Mak et al. / MRI Volumetry & ASL in AD and Normal Controls

same as the combination of all four parameters. Our383

multi-parametric MRI measures were analogous to384

previous results of improved discrimination of AD385

from CN by multimodal imaging (MRI plus PET)386

[13, 20].387

To assess whether the combined parameters model388

exhibited a higher diagnostic efficacy than the single389

parameter models, we tested the AUC MRL model390

(0.944) against each of the single parameter mod-391

els (i.e., AUC M (0.749), AUC P (0.744), AUC R392

(0.928), AUC L (0.872)) using one-sided Z-tests to see393

if any statistically significant difference existed. Given394

our limited sample size, only significant improve-395

ment of AUC MRL over AUC M (Z-score = 2.024,396

p = 0.022) and AUC P (Z-score = 2.062, p = 0.020) can397

be found. The effect of AUC MRL over AUC R398

or L remains to be verified. Nevertheless, a higher399

AUC MRL than AUC R or L is observed, and a nar-400

rower 95% CI for the AUC MRL is obtained when401

comparing with those of AUC R or L.402

Fifth, similar to prior studies [22, 23], we found in403

our cohort (AD plus CN) that the severity of neuropsy-404

chological impairments (MMSE, ADAS-cog, and SI)405

was strongly correlated with brain scan measurements.406

Using multiple linear regressions, the model combin-407

ing MC CBF and normalized RH and LH volumes408

was found to be a powerful predictor of the clinically409

observed severity measurements. Hence, our findings410

confirmed that the diagnostic probability of AD, its411

phenotypes, and cognitively normal elderly controls412

lie along a continuum and the brain scan measurements413

added certainty to the diagnostic process.414

Finally, in real-life clinical situation, we found that415

hippocampal volumetry is practical in differentiating416

early AD from normal aging as it can be implemented417

within 20 minutes in experienced hands [4]. The cut-418

off values of normalized RH and LH volumes in our419

Chinese cohort could be employed as reference for420

differentiation of the two groups. However, MRI struc-421

tural changes in AD occurred later than FDG-PET422

changes [24]. In combining both techniques in the423

same setting, the additional discriminatory effect of424

decreased CBF can be exploited for improvement in425

radiologic diagnosis [23]. The color-coded maps of426

CBF of a normal control and an AD patient (being427

an average in each diagnostic group in our study) were428

shown (Fig. 5), and the quality of the ASL MRI images429

allowed visual (qualitative) evaluation in the clinical430

setting [18].431

The major limitation of our study is a relatively432

small cohort. Also, we were unable to demonstrate433

in our cohort the superior sensitivity of the parietal434

Fig. 5. The CBF color-coded maps of a healthy control (72F) andan AD patient (78F, MMSE = 15), selected at two typical transversesections: level of basal ganglia and level of centrum semiovale.

association cortices in detecting CBF abnormalities, 435

as published in prior studies [22, 23]. Our unpub- 436

lished data (scatter-plot of CBF in the precuneus 437

of AD and healthy controls plotted against the SI 438

and MMSE) showed that a few controls demonstrate 439

impaired precuneus CBF, and one even had lower 440

CBF than AD subjects. These outliers, probably due 441

to technical CBF measurement variability or early 442

AD, might explain the insignificant result of precuneus 443

CBF. The pulsatile ASL technique has the drawback 444

of limited signal-to-noise, thereby affecting data qual- 445

ity and required cleaning in a recent study [25]. The 446

new pseudo-continuous approach holds promise of a 447

high efficiency and broad compatibility MRI perfusion 448

imaging method [23, 26]. Future large-scale clinical 449

studies are awaited to validate the accuracy and robust- 450

ness of the pulsed continuous ASL technique. 451

In summary, combined MRI using MRI hippocam- 452

pal volumetry and ASL CBF was superior to either 453

single measure in discriminating a cohort of AD and 454

CN. ASL has emerged as a competitive technique for 455

quantitative measurement of regional cerebral perfu- 456

sion. Being non-invasive and non-ionizing, there is a 457

critical need to establish the role of ASL in diagnosing 458

AD and its phenotypes. Since it can be combined with 459

Unc

orre

cted

Aut

hor P

roof

H.K.-F. Mak et al. / MRI Volumetry & ASL in AD and Normal Controls 9

routine MRI, a ‘one-stop-shop’ study of both structural460

and functional parameters will be highly desirable for461

both diagnosis and monitoring of disease-modifying462

therapies.463

DISCLOSURE STATEMENT464

Authors’ disclosures available online (http://www.j-465

alz.com/disclosures/view.php?id=2166).466

REFERENCES467

[1] Fayed N, Modrego PJ, Salinas GR, Gazulla J (2012) Magnetic468

resonance imaging based research in Alzheimer’s disease.469

J Alzheimers Dis 31, S5-S18.470

[2] Scheltens P, Fox N, Barkhof F, De Carli C (2002) Structural471

magnetic resonance imaging in the practical assessment of472

dementia: Beyond exclusion. Lancet Neurol 1, 13-21.473

[3] Dubois B, Feldman HH, Jacova C, DeKosky ST, Barberger-474

Gateau P, Cummings J, Delacourte A, Galasko D, Gauthier S,475

Jicha G, Meguro K, O’Brien J, Pasquier F, Robert P, Rosser476

M, Salloway S, Stern Y, Visser P, Scheltens P (2007) Research477

criteria for the diagnosis of Alzheimer’s disease: Revis-478

ing the NINCDS-ADRDA criteria. Lancet Neurol 6, 734-479

746.480

[4] Mak HKF, Zhang Z, Yau KKW, Zhang L, Chan Q, Chu481

LW (2011) Efficacy of voxel-based morphometry with DAR-482

TEL and standard registration as imaging biomarkers in a483

cohort of Alzheimer’s patients and cognitively normal older484

adults at 3.0 Tesla MR Imaging. J Alzheimers Dis 23, 655-485

664.486

[5] Glodzik-Sobanska L, Rusinek H, Mosconi L, Li Y, Zhan J,487

de Santi S, Convit A, Rich K, Brys M, de Leon MJ (2005)488

The role of quantitative structural imaging in the early diag-489

nosis of Alzheimer’s disease. Neuroimag Clin N Am 15, 803-490

826.491

[6] Petersen ET, Lim T, Golay X (2006) Model-free arterial spin492

labeling quantification approach for perfusion MRI. Magn493

Reson Med 55, 219-232.494

[7] Alsop DC, Detre JA, Grossman M (2000) Assessment of495

cerebral blood flow in Alzheimer’s disease by spin-labeled496

magnetic resonance imaging. Ann Neurol 47, 93-100.497

[8] Johnson NA, Jahng G, Weiner MW, Miller BL, Chui HC,498

Jagust WJ, Gorno-Tempini ML, Schuff N (2005) Pattern of499

cerebral hypoperfusion in Alzheimer disease and mild cog-500

nitive impairment measured with arterial spin-labeling MR501

imaging: Initial experience. Radiology 234, 851-859.502

[9] Chao LL, Pa J, Duarte A, Schuff N, Weiner MW, Kramer503

JH, Miller BL, Freeman KM, Johnson JK (2009) Patterns504

of cerebral hypoperfusion in amnesic and dysexecutive MCI.505

Alzheimer Dis Assoc Disord 23, 245-252.506

[10] Yoshiura T, Hiwatashi A, Noguchi T, Yamashita K, Ohyagi507

Y, Monji A, Nagao E, Kamano H, Togao O, Honda H (2009)508

Arterial spin labelling at 3-T MR imaging for detection of509

individuals with Alzheimer’s disease. Eur Radiol 19, 2819-510

2825.511

[11] Mak HKF, Chan Q, Zhang Z, Petersen ET, Qiu D, Zhang L,512

Yau KKW, Chu LW, Golay X (2012) Quantitative assessment513

of cerebral hemodynamic parameters by QUASAR arterial514

spin labeling in Alzheimer’s disease and cognitively normal515

elderly adults at 3-tesla. J Alzheimers Dis 31, 33-44.516

[12] Karow DS, McEvoy LK, Fennema-Notestine C, Hagler 517

DJ, Jennings RG, Brewer JB, Hoh CK, Dale AM, for the 518

Alzheimer’s Disease Neuroimaging, Initiative (2010) Rela- 519

tive capability of MR imaging and FDG PET to depict changes 520

associated with prodromal and early Alzheimer disease. Radi- 521

ology 256, 932-942. 522

[13] Jhoo JH, Lee DY, Choo IH, Seo EH, Oh JS, Lee JS, Lee 523

DS, Kim SG, Youn JC, Kim KW, Woo JI (2010) Discrimina- 524

tion of normal aging, MCI and AD with multimodal imaging 525

measures on the medial temporal lobe. Psychiatry Res 183, 526

237-243. 527

[14] Walhovd KB, Fjell AM, Amlien I, Grambaite R, Stenset 528

V, Bjornerud A, Reinvang I, Gjerstad L, Cappelen T, Due- 529

Tonnessen P, Fladby T (2009) Multimodal imaging in mild 530

cognitive impairment: Metabolism, morphometry and diffu- 531

sion of the temporal-parietal memory network. NeuroImage 532

45, 215-223. 533

[15] Chu LW, Tam S, Lee PWH, Yik PY, Song YQ, Cheung BMY, 534

Lam KSL (2009) Late-life mass index and waist circumfer- 535

ence in amnesic mild cognitive impairment and Alzheimer’s 536

disease. J Alzheimers Dis 17, 223-232. 537

[16] McKhann G, Drachman D, Folstein M, Katzman R, Price 538

D, Stadlan EM (1984) Clinical diagnosis of Alzheimer’s dis- 539

ease: Report of the NINCDS-ADRDA work group under the 540

auspices of Department of Health and Human Services Task 541

Force on Alzheimer’s Disease. Neurology 34, 939-944. 542

[17] Petersen ET, Mouridsen K, Golay X and on behalf of all 543

named co-authors of the QUASAR test-retest study (2010) 544

The QUASAR reproducibility study, Part II: Results from a 545

multi-center arterial spin labeling test-retest study. Neuroim- 546

age 49, 104-113. 547

[18] Raji CA, Lee C, Lopez OL, Tsay J, Boardman JF, Schwartz 548

ED, Bartynski WS, Hefzy HM, Gach HM, Dai W, Becker 549

JT (2010) Initial experience in using continuous arterial spin- 550

labeled MR imaging for early detection of Alzheimer disease. 551

AJNR Am J Neuroradiol 31, 847-855. 552

[19] De Santi S, de Leon MJ, Rusinek H, Convit A, Tarshish 553

CY, Roche A, Tsui WH, Kandil E, Boppana M, Daisley K, 554

Wang GJ, Schlyer D, Fowler J (2001) Hippocampal forma- 555

tion glucose metabolism and volume losses in MCI and AD. 556

Neurobiol Aging 22, 529-539. 557

[20] Kawachi T, Ishii K, Sakamoto S, Sasaki M, Mori T, Yamashita 558

F, Matsuda H, Mori E (2006) Comparison of the diagnostic 559

performance of FDG-PET and VBM-MRI in the very mild 560

Alzheimer’s disease. Eur J Nucl Med Mol Imaging 33, 801- 561

809. 562

[21] Walhovd KB, Fjell AM, Brewer J, McEvoy LK, Fennena- 563

Notestine C, Hagler DJ, Jenning RG, Karow D, Dale AM, 564

and the Alzheimer’s Disease Neuroimaging Initiative (2010) 565

Combining MR imaging, positron-emission tomography, and 566

CSF biomarkers in the diagnosis and prognosis of Alzheimer 567

disease. AJNR Am J Neuroradiol 31, 347-354. 568

[22] Weston Ashford J, Shih W, Coupal J, Shetty R, Schneider 569

A, Cool C, Aleem A, Kiefer VH, Mendiondo MS, Schmitt 570

FA (2000) Single SPECT measures of cerebral cortical per- 571

fusion reflect time-index estimation of dementia severity in 572

Alzheimer’s disease. J Nucl Med 41, 57-64. 573

[23] Binnewijzend MAA, Kuijer JPA, Benedictus MR, van der 574

Flier WM, Meije Wink A, Wattjes MP, van Berckel BNM, 575

Scheltens P, Barkhof F (2013) Cerebral blood flow measured 576

with 3D pseudocontinuous arterial spin-labeling MR imag- 577

ing in Alzheimer disease and mild cognitive impairment: A 578

marker for disease severity. Radiology 267, 221-230. 579

[24] Jack CR Jr, Knopman DS, Jagust WJ, Shaw LM, Aisen 580

PS, Weiner MW, Petersen RC, Trojanowski JQ (2010) 581

Unc

orre

cted

Aut

hor P

roof

10 H.K.-F. Mak et al. / MRI Volumetry & ASL in AD and Normal Controls

Hypothetical model of dynamic biomarkers of the582

Alzheimer’s pathological cascade. Lancet Neurol 9, 119-128.583

[25] Wang Z, Das SR, Xie SX, Arnold SE, Detre JA, Wolk DA,584

Alzheimer’s Disease Neuroimaging Initiative (2013) Arterial585

spin labeled MRI in prodromal Alzheimer’s disease: A multi-586

site study. Neuroimage Clin 2, 630-636.587

[26] Dai W, Garcia D, de Bazelaire C, Alsop DC (2008) Con- 588

tinuous flow-driven inversion for arterial spin labeling using 589

pulsed radio frequency and gradient fields. Magn Reson Med 590

60, 1488-1497. 591