Visual rating method and tensor-based morphometry in the diagnosis of mild cognitive impairment and Alzheimer's disease: a comparative magnetic resonance imaging study

Abstract

Background

Atrophy of the medial temporal lobe (MTL) is the main structural magnetic resonance imaging (MRI) finding in the brain of patients with Alzheimer's disease (AD). However, evaluating the degree of atrophy is still demanding.

Purpose

The visual rating method (VRM) was compared with multi-template tensor-based morphometry (TBM), in terms of its efficacy in diagnosing of mild cognitive impairment (MCI) and AD.

Material and Methods

Forty-seven patients with MCI, 80 patients with AD and 84 controls were studied.

Results

TBM seems to be more sensitive than VRM at the early stage of dementia in the areas of MTL and ventricles. The methods were equally good in distinguishing controls and the MCI group from the AD group. At the frontal areas TBM was better than VRM in all comparisons.

Conclusion

A user-friendly VRM is still useful for the clinical evaluation of MCI patients, but multi-template TBM is more sensitive for diagnosing the early stages of dementia. However, TBM is currently too demanding to use for daily clinical work.

Keywords

Visual rating tensor-based morphometry Alzheimer's disease mild cognitive impairment magnetic resonance imaging (MRI)

Introduction

Atrophy of the medial temporal lobe (MTL) and hippocampal atrophy in particular are associated with the severity of Alzheimer's disease (AD) (1,2). The atrophy is also progressive and spreads throughout the brain to cause multiform dysfunction, such as impairment in executive functions when reaching the frontal lobes (2 –4). Due to parenchymal shrinking, the ventricles widen (5). Ventricular enlargement in mild cognitive impairment (MCI) and AD is stronger than that observed in normal aging and is associated with cognitive decline (2,6).

The structural magnetic resonance imaging (MRI) based visual rating method (VRM) was developed to evaluate the medial temporal atrophy (MTA) (7). The VRM can separate AD, MCI, and other dementia subjects from healthy controls (1,2,7). VRM is easy to learn and does not require sophisticated software, and the analyses can be performed cost-effectively. However, the usability of VRM is limited in estimating the atrophy of brain areas other than MTA.

Tensor-based morphometry (TBM) is a fully automated voxel-based method for identifying and quantifying structural differences in the brain (8). It is also free of subjective interpretations and the analysis extends outside the MTL. A multi-template TBM is usable at very early stages of dementia and it separates controls, MCI patients and AD patients from each other most clearly in MTL and the hippocampus but also in several other brain structures (9,10). The limitations for its clinical use are the need for extra time, special expertise, and computers with high capacity.

The aim of the present study was to compare the VRM of MTA, ventricular enlargement, and frontal atrophy to measures of atrophy in respective brain areas using multi-template TBM. We hypothesized that TBM would be a more sensitive method for revealing brain atrophy in MCI and AD compared with any of the VRMs.

Material and Methods

Subjects

Study participants were recruited at Turku University Hospital (PET Centre and Department of Neurology). The study included patients with MCI, patients with AD, and healthy controls. MCI was diagnosed according to the criteria suggested by Petersen et al. (11). Patients with AD fulfilled the DSM-IV criteria for dementia and the NINCS-ADRDA (National Institute of Neurological and Communicative Disorders and Stroke/Alzheimer's Disease and Related Disorders Association) criteria for probable AD (12). The control subjects were healthy volunteers who had no history of neurological or psychiatric disease and had scores within the age-adjusted Finnish norms in neuropsychological testing. Altogether, 211 subjects (47 subjects with MCI, 80 with AD and 84 controls) were included in the study. MRIs of the subjects' brain were analyzed using both the VRM and TBM. In the TBM study there were 47 subjects in the MCI group, 79 in the AD group, and 83 in the control group.

The study was approved by the Joint Ethical Committee of the University of Turku and Turku University City Hospital and performed in accordance with the 1964 Declaration of Helsinki and its later amendments. The subjects received oral and written information about the study and gave informed consent prior to their inclusion in this study.

Neuropsychological testing

The neuropsychological tests were used to ensure normal cognition in controls and the absence of widespread cognitive impairment in MCI patients and included CERAD (13), the Logical Memory test from the Revised Wechsler Memory Scale (14), and the Trail Making Test (15). These tests evaluated instance episodic memory, language function, visuospatial function, and executive function.

Magnetic resonance imaging

The MRI scanning of the subjects was performed with either a 1.5T Philips Intera (Best, The Netherlands) or a 1.5 T MRI GE Signa Horizon LX EchoSpeed (GE Healthcare, Milwaukee, WI, USA). The Philips scanner used an axial three-dimensional (3D) T1/FFE sequence for analysis (voxel size, 0.50 × 0.50 × 1.00 mm). The GE scanner used an axial 3D FSPGR sequence (voxel size, 1.1 × 1.1 × 1.5 mm).

For the visual evaluation the axial T1 3D sequences were reconstructed manually at a GE workstation (ADW 4.4, GE Healthcare). During reconstruction the slices were aligned with a brainstem axis (7). The interslice gap was 1.0 mm and the slice thickness was 5.0 mm. Frontal atrophy evaluation was performed from the original sequences. The evaluation of the MTA and ventricular enlargement was performed by a radiologist and a neuroradiologist and the evaluation of the frontal atrophy was performed by a neuroradiologist. The evaluators were blinded to the subjects' data and the order of MRI scans was randomized. The radiologists analyzed the sequences independently, and differences of more than one grade were resolved by consensus; the higher value was retained for ratings differing by only one grade.

Visual rating method

MTL evaluation in the VRM was performed as described by Scheltens et al. (7). Atrophy was scaled from value 0 to 4, where a value of 0 corresponded to no atrophy and a value of 4 corresponded to severe atrophy. Absolute measurements (mm) of the largest vertical height of hippocampal formation (defined as dentate gyrus, hippocampus proper, subiculum, and parahippocampal gyrus), the largest horizontal width between the hippocampal formation and brainstem, the largest vertical width of the choroid fissure, and the width of temporal horn were calculated. The absolute measurements of the largest width of the lateral ventricles and third ventricle and the diameter of the inner calvarium of the skull were also measured in mm and from the same slice as the MTL measurements. To calculate the indices, all of the absolute measurements of MTL, the surrounding cerebrospinal fluid (CSF) spaces and ventricles were proportioned to an absolute diameter of an inner calvarium of the skull. The aim of the indices was to create results that would be independent of the size of the subject's skull.

Frontal atrophy evaluation was based on a study by Victoroff et al. (16). The evaluation was scaled in the range of 0–3, where 0 corresponded to no atrophy and 3 corresponded to severe atrophy.

Tensor-based morphometry

The T1 3D MRI sequences were also used for the TBM analysis. A multi-template method of TBM was used to improve the robustness of the analysis (10). The template images used in this study were a compilation of 10 AD subjects, 10 MCI subjects, and 10 controls from the ADNI dataset (http://adni.loni.usc.edu/). A mean anatomical template (MAT) was constructed from the 30 templates to establish a reference space for the analysis.

For each study image, 30 separate registrations from MAT to the study image were computed. Each time, the registration was performed via a different template. For each registration, the determinants of the Jacobian matrix, (Jacobians) were computed for each voxel to quantify the local volume changes. The 30 deformations were combined by averaging the Jacobians for each voxel. The resulting average Jacobian described the amount of local expansion or compression compared with the MAT.

The average Jacobians were further combined within each structure of the Hammer's brain atlas (17) to produce robust values for further analysis. The final TBM index was computed for each structure by measuring the similarity of the subject's Jacobians with the typical AD-related pattern of Jacobians modelled from the ADNI dataset as described in detail by Koikkalainen et al. (10). High index values indicated AD-type shape changes from the MAT, whereas low values showed similarity to the shape changes of typical controls.

Statistical analysis

The groups were compared with each other by statistical analysis. The groups were significantly different with respect to age and education. Age differences were analyzed by one way ANOVA, education differences were analyzed by the Kruskal-Wallis test and Mann-Whitney U test. ANCOVA and the cumulative logistic regression models were used to analyze possible main effects of the groups. The effects of age and education were used as fixed effects. In the ANCOVA model, a difference in the estimated means (DEM) was used in the visual rating analysis and a difference in the estimated index values (DEIV) was used in the TBM analysis. A negative DEM or DEIV means that in the comparison between groups, the latter group has higher average values than the first group does: positive DEM or DEIV values indicate the opposite. In the analysis of visual frontal atrophy, RO is a coefficient ratio of the estimated means in the regression model. If the RO value is higher than one, it is more likely that in the comparison between groups the first group has higher values than the second group. If RO is smaller than one, the interpretation is opposite. The result of the comparison is statistically non-significant if the 95% confidence interval contains the value 1. All results are reported as corrected for multiple comparisons by using Tukey's P values.

To evaluate how well the estimation of hippocampal atrophy by VRM and TBM managed to separate the groups from each other, analysis of the receiver operating characteristic (ROC) was performed. The area under the ROC curve (AUC) shows how well the test differentiates the groups. The AUC values are generally characterized as excellent (0.9–1.0), good (0.8–0.9), fair (0.7–0.8), poor (0.6–0.7), and failed (0.5–0.6).

Analyses were performed using SAS 9.2 (SAS Institute Inc., Cary, NC, USA).

Results

Demographic and neuropsychological characterizations

Demographic details of the study subjects are shown in Table 1. Neuropsychological characteristics of study subjects are presented in Table 2.

Table 1.

Demographic details of different study groups.

Demography	Controls	MCI	AD	Stat. sign.*
Number of subjects (211) (109 men, 102 women)	84	47	80
Sex male/female	41/43	27/20	41/39	n.s.
Age (years), mean (SD)	71.0 (6.7)	72.9 (6.3)	74.6 (5.5)	AD > C
– Median	71.4	72.6	73.7
– Min/Max	51.8/87.8	59.0/85.5	57.3/85.9
Education (^a), mean (SD)	1.9 (0.9)	2.0 (1.2)	1.6 (0.9)	C, MCI > AD
– Median	2	2	1
– Min/Max	1/4	1/4	1/4
MMSE, mean (SD)	27.8 (0.35)	26.5 (0.48)	22.3 (0.41)	C>MCI, AD; C, MCI > AD
– Median	28	26	23
– Min/Max	25/30	25/30	6/29

α < 0.05.

Education (^a) level was operationalized as follows: 1, comprehensive school; 2, vocational school; 3, college degree; and 4, university degree. Statistical details in text in paragraph “Statistical analysis”.

Table 2.

Results on the neuropsychological tests.

Analysis of neuropsychological tests	Controls	MCI subjects	AD subjects	Controls vs. MCI subjects	Controls vs. AD subjects	MCI vs. AD subjects
	EM (SD)	EM (SD)	EM (SD)	DEM	DEM	DEM
MMSE	27.8 (0.35)	26.5 (0.48)	22.3 (0.41)	1.356*	5.497†	4.142†
Neuropsychological test
Episodic memory
- Logic memory	20.8 (1.37)	16.4 (1.61)	12.0 (1.55)	4.351‡	8.745†	4.394‡
- Logic memory delayed	17.0 (1.43)	11.5 (1.68)	7.4 (1.62)	5.477‡	9.612†	4.135*
- Wordlist learning (sum)	20.2 (1.00)	19.1 (1.17)	15.2 (1.14)	1.104	5.054†	3.951§
- Word delayed recall	7.3 (0.44)	6.1 (0.52)	4.4 (0.50)	1.288*	2.933†	1.645‡
- Wordlist savings (%)	91.8 (5.98)	82.8 (7.04)	63.3 (6.76)	9.033	28.549†	19.517‡
- Wordlist recogn. (%) (^b)	97.0 (2.04)	92.9 (2.40)	86.7 (2.31)	4.105	10.292†	6.187*
- Constr. praxis savings (^c)	7645.0 (682.61)	7327.5 (801.66)	5608.5 (779.82)	317.540	2036.520‡	1718.990
Verbal functions
- Semantic fluency	22.7 (1.29)	19.9 (1.51)	19.6 (1.46)	2.746	3.052*	0.305
- Naming	13.3 (0.45)	12.6 (0.53)	10.9 (0.51)	0.673	2.370†	1.697‡
Visuospatial function
- Constructional praxis (^b)	9.9 (0.37)	10.1 (0.43)	9.5 (0.42)	−0.197	0.375	0.572
Executive functions
- Clock drawing	5.4 (0.31)	5.2 (0.36)	4.1 (0.35)	0.159	1.287†	1.128‡
- Trail making A	71.5 (8.43)	81.4 (9.85)	97.4 (9.36)	−9.850	−25.854‡	−16.005
- Trail making B	142.6 (22.21)	172.1 (27.73)	222.6 (26.38)	−29.547	−80.067‡	−50.520

α < 0.05.

†

α < 0.0001.

‡

α < 0.01.

α < 0.001.

Distribution of Constructional praxis and Wordlist recognition (%) task values were highly skewed (^b) and they were not able to be converted normal distribution. Superscript (^c) means a square root conversion.

DEM, difference of the estimated means; EM, estimate of the means; SD, standard deviation.

Visual rating method

There were statistically significant differences between the controls and the AD group with respect to the medial temporal lobe and ventricular width (Table 3). The negative DEM indicated that the AD group had higher atrophy scores in the hippocampus, wider CSF spaces around the hippocampus and larger ventricles than did the control group. In turn, the positive DEM showed that the AD group had a lower height of the hippocampus and parahippocampal gyrus, which was in line with the results of hippocampal scoring.

Table 3.

Results of the visual rating analysis.

Visual rating analysis		Controls	MCI subjects	AD subjects	Controls vs. MCI subjects	Controls vs. AD subjects	MCI vs. AD subjects
		EM (SD)	EM (SD)	EM (SD)	DEM	DEM	DEM
Medial temporal lobe
– Score	Right	1.1 (0.12)	1.3 (0.16)	2.2 (0.14)	−0.143	−1.056*	−0.912*
	Left	1.2 (0.13)	1.4 (0.18)	2.2 (0.15)	−0.251	−1.038*	−0.786†
– Indexes
A. Height of hippocampus and	Right	0.139 (0.0021)	0.139 (0.0028)	0.127 (0.0024)	0.000	0.012*	0.012‡
Parahippocampal gyrus	Left	0.138 (0.0020)	0.134 (0.0027)	0.128 (0.0023)	0.004	0.011†	0.007
Liquor space of temporal horns
– Indexes
B. Distance between hippocampus and	Right	0.031 (0.0012)	0.031 (0.0016)	0.034 (0.0013)	−0.001	−0.004‡	−0.003
Brainstem	Left	0.029 (0.0011)	0.029 (0.0014)	0.034 (0.0012)	0.000	−0.005§	−0.004
C. Width of choroid fissure	Right	0.021 (0.0012)	0.022 (0.0017)	0.029 (0.0014)	−0.001	−0.009*	−0.008†
	Left	0.021 (0.0011)	0.021 (0.0015)	0.027 (0.0013)	0.000	−0.006*	−0.006†
D. Width of temporal horn	Right	0.019 (0.0017)	0.019 (0.0022)	0.030 (0.0019)	0.000	−0.010*	−0.010†
	Left (^d)	−3.962 (0.0675)	−3.829 (0.0900)	−3.609 (0.0767)	−0.132	−0.353†	−0.221
E = B + C + D	Right	0.071 (0.0032)	0.072 (0.0043)	0.093 (0.0036)	−0.001	−0.022*	−0.021*
	Left	0.072 (0.0033)	0.075 (0.0044)	0.094 (0.0037)	−0.004	−0.022*	−0.018†
Ventricles
– Indexes
Width of lateral ventricles		0.280 (0.0053)	0.294 (0.0071)	0.300 (0.0060)	−0.013	−0.020‡	−0.007
Width of third ventricle		0.055 (0.0022)	0.063 (0.0030)	0.067 (0.0026)	−0.008‡	−0.012†	−0.004
Frontal atrophy		Mean (SD)	Mean (SD)	Mean (SD)	RO	RO	RO
– Score	Right	0.52 (0.77)	0.83 (1.14)	0.64 (0.92)	0.827	1.138	0.941
	Left	0.51 (0.77)	0.83 (1.14)	0.64 (0.92)	0.538	1.208	0.651

α < 0.0001.

†

α < 0.001.

‡

α < 0.05.

α < 0.01.

Superscript (^d) means conversion of logarithm.

DEM, difference of the estimated means; EM, estimate of the means; RO, ratio of coefficient; SD, standard deviation.

VRM also separated the MCI group from the AD group in most of the areas of the medial temporal lobe but not in the area of the ventricles. However, the positive DEM of the height of the left hippocampus and parahippocampal gyrus indicated that the control group had a higher left hippocampus value than the MCI group. A similar trend was observed on the right side.

Statistical differentiation of the MCI group from controls was only achieved by the greater width of the third ventricle in the MCI group. However, a trend of greater atrophy in the areas of medial temporal lobe and ventricles in the MCI group compared with the control group (a negative DEM) was observed.

In the frontal atrophy scores there were no significant differences between the study subject groups in any of the comparisons.

Tensor-based morphometry

Table 4 shows estimates of the TBM index values (EIVs) for the control, MCI and AD groups. High EIVs indicate similarity to AD, and low EIVs indicate similarity to control subjects. TBM performed the best at differentiating the control group from the AD group in the areas of the medial temporal lobe and ventricles, but the differentiation was also good in the frontal area. The higher EIVs in the AD group compared with the control group (negative DEIV) shows a better fit to the AD type volume changes on average.

Table 4.

Results of the TBM analysis.

Tensor-based morphometry analysis		Controls	MCI	AD	Controls vs. MCI subjects	Controls vs. AD subjects	MCI vs. AD subjects
		EIV	EIV	EIV	DEIV	DEIV	DEIV
Medial temporal lobe
A. Hippocampus	Right	0.0249	0.0743	0.2149	−0.049	−0.190*	−0.141†
	Left (^c)	0.4929	0.5747	0.6692	−0.082‡	−0.176*	−0.095†
B. Anterior temporal lobe,	Right	0.1267	0.1636	0.2287	−0.037	−0.102§	−0.065
medial part	Left (^d)	−0.5079	−0.4635	−0.3728	−0.044	−0.135§	−0.091
C. Parahippocampal gyrus and	Right	0.1111	0.1510	0.1768	−0.040	−0.066§	−0.026
ambiens	Left	0.1307	0.1897	0.2552	−0.059‡	−0.124*	−0.065§
D = A + B + C	Right	0.0607	0.1203	0.2755	−0.060	−0.215*	−0.155†
	Left (^c)	0.5359	0.6132	0.7198	−0.077§	−0.184	−0.107†
Ventricles
A. Lateral ventricle (frontal and	Right	−0.0339	0.0834	0.1946	−0.1173	−0.2285§	−0.1111
occipital horn and central part)	Left	−0.1795	−0.0945	0.0546	−0.0850	−0.2342§	−0.1492
B. Lateral ventricle (temporal horn)	Right	−0.0323	−0.0268	0.1096	−0.0055	−0.1420*	−0.1364†
	Left	0.0432	0.0958	0.2131	−0.0526	−0.1698*	−0.1172†
C. Third ventricle		−0.0430	0.1050	0.0999	−0.1480‡	−0.1429†	0.0050
D = A + B + C	Right	−0.0309	0.0839	0.1960	−0.1148	−0.2269§	−0.1121
	Left	−0.1733	−0.0888	0.0623	−0.0845	−0.2356§	−0.1512
Frontal area (prefrontal)
- Frontal lobe (Middle frontal gyrus)	Right	0.1862	0.1884	0.2417	−0.0022	−0.0555‡	−0.0533
	Left	0.2152	0.2153	0.2837	−0.0001	−0.0685‡	−0.0684‡
- Superior frontal gyrus	Right (^c)	0.4828	0.5346	0.5162	−0.0518‡	−0.0334	0.0185
	Left	0.2217	0.2599	0.2505	−0.0382	−0.0288	0.0094
- Straight gyrus	Right	−0.0558	−0.0565	−0.0136	0.0007	−0.0422‡	−0.0429
	Left	−0.0579	−0.0837	−0.0098	0.0258	−0.0481‡	−0.0740†
- Medial orbital gyrus	Right	−0.0628	−0.0734	−0.0122	0.0107	−0.0506†	−0.0613†
	Left	0.0134	−0.0244	0.0699	0.0378	−0.0565‡	−0.0943§
- Posterior orbital gyrus	Right	0.0144	0.0292	0.0623	−0.0148	−0.0479	−0.0331
	Left	0.1057	0.1204	0.1673	−0.0147	−0.0616†	−0.0469
Global (the whole brain)		0.0080	0.1051	0.2459	−0.0971	−0.2379*	−0.1408‡

α < 0.0001.

†

α < 0.01.

‡

α < 0.05.

α < 0.001.

Superscript (^c) means square root conversion and (^d) conversion of logarithm.

DEIV, difference of the estimated index values; EIV, estimated index values.

When the MCI and AD groups were compared, only the hippocampus, the combined areas of MTL, temporal horns of the lateral ventricles, and medial orbital gyri of the frontal area gave statistically significant results on both sides. The DEIVs were also negative. The results were also statistically significant on the left side of the parahippocampal gyrus and ambiens and in the middle frontal and straight gyri of the frontal area.

When the controls and the MCI group were compared, statistically significant results of the MTL were only observed on the left side of hippocampus, parahippocampal gyrus, and ambiens and the combined values of the MTL. In the area of the ventricles, the third ventricle had the only statistically significant P value. In the frontal area the only statistically significant area was in the right side of the superior frontal gyri. When the DEIV was examined, the negative values in the areas of MTL, ventricles, and the superior frontal gyri meant that, on average, the MCI group had more AD type volume changes than did the control group.

It is possible to visualize the volume changes by TBM. Fig. 1 demonstrates the differences between the groups by color.

Fig. 1.

The visualization of the TBM results for individual AD, MCI, and control subjects. The visualization is based on the z-scores of the voxel-wise Jacobians. The z-scores were computed by comparing the individual data to the entire control dataset. The red color shows the regions with AD-type brain dilation and blue color shows the regions with AD-type brain atrophy. The AD-type brain change (i.e. locations where dilation/atrophy occurs) was computed from the AD and controls datasets. Only the regions with statistically (P < 0.05) significant differences between AD and control groups, z-scores larger than 1.0 or smaller than –1.0, are visualized in colors.

Receiver-operating characteristic analyses

ROC AUC analyses were performed to compare whether there was a sufficient difference in the hippocampal atrophy by VRM and TBM to differentiate the groups from each other. The AUCs were not significantly different between the groups for any of the variables of interest, and the values of the AUCs were in the range of 0.564–0.768 (Table 5). In the left side, the AUC of TBM was larger than the VRM at a trend level (P = 0.09) when the controls were compared with the MCI subjects. In the right side, as for the AUC of the VRM, they were larger than TBM at a trend level (P = 0.07) when the MCI subjects were compared with the AD subjects.

Table 5.

ROC analysis.

ROC	Controls vs. MCI subjects				Controls vs. AD subjects				MCI subjects vs. AD subjects
	TBM	VRM	Estimate	P value	TBM	VRM	Estimate	P value	TBM	VRM	Estimate	P value
Right side	0.6417	0.5642	0.0775	0.15	0.7407	0.7680	−0.0273	0.40	0.6367	0.7066	−0.0699	0.07
Left side	0.6796	0.5877	0.0919	0.09	0.7774	0.7609	0.0165	0.63	0.6510	0.6782	−0.0272	0.51

Comparison analysis of AUCs between the hippocampus of TMB and the hippocampus score of VRM. The hippocampus score of VRM was used as a reference.

Estimate, difference of AUCs between TBM and VRM.

Discussion

When comparing VRM and TBM and estimating their utility in differentiating patients with MCI and AD from healthy controls, our two main findings were as follows: (i) when comparing the controls with the MCI group, TBM indicated reduced left sided hippocampal volume in MCI patients whereas the hippocampal atrophy scores derived using the VRM did not show statistically significant differences; and (ii) the hippocampal atrophy and ventricular enlargement separated the MCI group and controls from the AD group with both methods. In frontal areas statistically significant differences were found only by using TBM. In ROC analyses, there was only a trend that TBM of the left hippocampus differentiated MCI patients from controls better than VRM.

It was surprising that only TBM revealed statistically significant differences between controls and the MCI groups in the hippocampal atrophy although the mean hippocampal atrophy values on both sides were greater in the MCI group than in controls with both methods. Earlier studies have shown that patients with amnestic MCI have greater hippocampal atrophy scores in the VRM than do controls, thus predicting progressive MCI and conversion to AD (1,2,18). It should be kept in mind, however, that MCI is heterogeneous and that not all MCI subjects will progress to AD or other dementias.

When comparing controls and the MCI group with the AD group, atrophy of the MTL and ventricular enlargement had progressed in the AD group and both methods distinguished the groups from each other. However, controls differed from the AD group more clearly than did the MCI group from the AD group.

In frontal areas statistically significant differences were found only by using TBM. However, in the comparison between controls and the MCI group, a difference was only observed in one frontal area. In other comparisons, the frontal atrophy was more apparent. However, together with the non-significant difference between the groups in frontal atrophy in visual evaluation, we suggest that VRM is a less sensitive method than TBM.

There were some limitations in this study. The number of subjects in the groups was relatively low, particularly in the MCI group. From a methodological point of view, the VRM may be dependent on the experience of the evaluators (18). We tried to reduce the bias of evaluation by conducting the scoring according to the published guidelines and example images (7). The limitations of the multi-template TBM analysis are as follows: (i) additional computation time is required; (ii) the structures have been inaccurately segmented from the MAT; (iii) the precision of the registration algorithm is inaccurate; and (iv) the Jacobian does not measure any other form of shape change.

In conclusion, our findings show that multi-template TBM is more sensitive than VRM at the early stages of dementia. The difference is observed most clearly in the left hippocampal area. The methods are equally good in distinguishing controls and the MCI group from the AD group at the areas of the MTL and ventricles. VRM is insensitive for frontal atrophy, but at the early stages TBM is also not a very robust method. Based on our findings, a user-friendly VRM has utility for the clinical evaluation of MCI patients, but TBM is better at the most important early stage diagnostics. However, an easy user interface for the TBM analysis must be developed before its application in clinical work.

Footnotes

Acknowledgements

Sofia Männikkö, a biostatistician from Department of Biostatistics in Turku University, Finland, provided statistical expertise.

Conflict of interest

None declared.

Funding

The study was supported by the Academy of Finland (project no. 133193), the Sigrid Juselius Foundation, and Turku University Hospital Clinical Grants.

References

Duara

Loewenstein

Potter

. Medial temporal lobe atrophy on MRI scans and the diagnosis of Alzheimer's disease. Neurology 2008; 71: 1986–1992.

Frisoni

Fox

Jack

Jr . The clinical use of structural MRI in Alzheimer's disease. Nat Rev Neurol 2010; 6: 67–77.

Cardenas

Chao

Studholme

. Brain atrophy associated with baseline and longitudinal measures of cognition. Neurobiol Aging 2011; 32: 572–580.

Lindberg

Westman

Karlsson

. Is the subcallosal medial prefrontal cortex a common site of atrophy in Alzheimer's disease and frontotemporal lobar degeneration? Front Aging Neurosci 2012; 4: 32.

Madsen

Gutman

Joshi

. Mapping dynamic changes in ventricular volume onto baseline cortical surfaces in normal aging, MCI, and Alzheimer's disease. Multimodal Brain Image Anal 2013; 8159: 84–94.

Chou

Lepore

Saharan

. Ventricular maps in 804 ADNI subjects: Correlations with CSF biomarkers and clinical decline. Neurobiol Aging 2010; 31: 1386–1400.

Scheltens

Leys

Barkhof

. Atrophy of medial temporal lobes on MRI in “probable” Alzheimer's disease and normal ageing: Diagnostic value and neuropsychological correlates. J Neurol Neurosurg Psychiatry 1992; 55: 967–972.

Ashburner

Hutton

Frackowiak

. Identifying global anatomical differences: Deformation-based morphometry. Hum Brain Mapp 1998; 6: 348–357.

Wolz

Julkunen

Koikkalainen

. Multi-method analysis of MRI images in early diagnostics of Alzheimer's disease. PLoS One 2011; 6: e25446.

10.

Koikkalainen

Lotjonen

Thurfjell

. Multi-template tensor-based morphometry: Application to analysis of Alzheimer's disease. Neuroimage 2011; 56: 1134–1144.

11.

Petersen

Stevens

Ganguli

. Practice parameter: Early detection of dementia: Mild cognitive impairment (an evidence-based review). Report of the Quality Standards Subcommittee of the American Academy of Neurology. Neurology 2001; 56: 1133–1142.

12.

McKhann

Drachman

Folstein

. Clinical diagnosis of Alzheimer's disease: Report of the NINCDS-ADRDA work group under the auspices of department of health and human services task force on Alzheimer's disease. Neurology 1984; 34: 939–944.

13.

Morris

Heyman

Mohs

. The consortium to establish a registry for Alzheimer's disease (CERAD). Part I. Clinical and neuropsychological assessment of Alzheimer's disease. Neurology 1989; 39: 1159–1165.

14.

Wechsler

. Wechsler memory scale-revised: Manual, San Antonio, TX: The Psychological Corporation & Harcourt Brace Jovanovich, Inc.;, 1987.

15.

Reitan

. A manual for the administration and scoring of the trail making test, Bloomington, IN: Indiana University Press, 1959.

16.

Victoroff

Mack

Grafton

. A method to improve interrater reliability of visual inspection of brain MRI scans in dementia. Neurology 1994; 44: 2267–2276.

17.

Heckemann

Hajnal

Aljabar

. Automatic anatomical brain MRI segmentation combining label propagation and decision fusion. Neuroimage 2006; 33: 115–126.

18.

Scheltens

Launer

Barkhof

. Visual assessment of medial temporal lobe atrophy on magnetic resonance imaging: Interobserver reliability. J Neurol 1995; 242: 557–560.