Grey-matter volume as a potential feature for the classification ofAlzheimer’s disease and mild cognitive impairment: An exploratory study

Yan’e Guo 1, Zengqiang Zhang 1,2, Bo Zhou 1, Pan Wang 1,Hongxiang Yao3, Minshao Yuan4, Ningyu An3, Haitao Dai2, Luning Wang 1, Xi Zhang1, Yong Liu5,6

1Department of Neurology, Institute of Geriatrics and Gerontology, ChinesePLAGeneralHospital, Beijing 100853, China

2Hainan Branch of ChinesePLAGeneralHospital, Sanya 572014, China

3Department of Radiology, ChinesePLAGeneralHospital, Beijing 100853, China

4Department of Neurology, the People’s Hospital of Jimo, Qingdao 266200, China

5BrainnetomeCenter, Institute of Automation, ChineseAcademy of Sciences, Beijing 100190, China

6National Laboratory of Pattern Recognition, Institute of Automation, ChineseAcademy of Sciences, Beijing 100190, China

Corresponding authors: Xi Zhang and Yong Liu. E-mail: ,

Fig. S1. Whole-brain voxel-wise analyses between groups. A-C: Brain areas with significant atrophy in aMCI patients compared to NC individuals (A); AD compared to aMCI patients (B); and AD patients compared to NC individuals (C). In all cases,P<0.001, cluster size 40 voxels, uncorrected.

Fig.S2. Subregion segmentation. The left hippocampus was isolated based on the AAL template using the WFU_PickAtlas toolkit () [1] and included 20 slices (at y-axis) from the head to the tail and 932 voxels (resolution 2 mm×2 mm×2 mm). Next, we divided these 20 slices into 4 equal parts. The voxel numbers for the head, body 1, body 2, and tail were 264, 202, 265, and 201, respectively. The right hippocampus and the bilateral parahippocampus were subdivided in the same manner. A and B illustrate the parcellations of the bilateral hippocampus and parahippocampus. C: The gray-matter volumes of these subdivided regions were significantly decreased in AD (grey bar) compared to NC groups (white bar). *P<0.05 (FDR-corrected).

1/6

Fig. S3. ROC curves of the leave-2-to-5-out analyses to classifyparticipants with AD and NC using the linear discrimination analysis method.A: We randomly removed 2 participants who formed the test group and examined the remaining 60 as the training group. This cross-validation step was performed 1770 (60×59/2) times. The average proportion of correct classifications between the AD and NC participants was 82.3%. B-D: Similarly, we randomly removed 3-to-5 participants as test datasets and performed 10,000 cross-validations; the mean correct ratio was ~83%. AUC, area under the curve; correct ratio, percentage of participants correctly classified out of the total number of participants.

Fig. S4. The same cross-validation procedure was used to distinguish the AD, aMCI, and NC groups. A-D: ROC curves showing that the leave-2-to-5-out analyses resulted in classification of NC(blue), aMCI(red), and AD(black) participants when linear discrimination analysis was used.AUC, area under the curve;correct ratio, percentage of participants correctly classified out of the total number of participants.

Fig. S5. Brain areas with significant grey matter atrophy in AD patients compared to NC individuals. A: left hemisphere. B: right hemisphere (P<0.001, FWE corrected, cluster size 40 voxels). Further details on these regions are in Table S1.

Table S1. Regions with significantly reduced gray-matter volume in AD patients compared to normal controls

Region / BA / Cluster size / T value / Z value / MNI Coordinates (x,y,z)
Hip/Phip.L / 28/34 / 362 / 10.94 / 8.34 / -28 -12 -16
7.38 / 6.17 / -20 -22 -14
6.84 / 5.83 / -26 2 -22
Hip/Phip.R / 28/34 / 386 / 9.43 / 7.31 / 26 -10 -16
7.51 / 6.25 / 38 -30 -12
STG/MTG.L / 21/22 / 246 / 8.96 / 7.07 / -56 -38 -4
Hip/Phip.R / 30 / 82 / 8.10 / 6.60 / 20 -36 0
Hip/Phip.L / 27 / 67 / 7.69 / 6.36 / -20 -36 -2

Two-sample,two-tailed t-tests with age and gender as covariants. The threshold for the resultant T-value map was chosen to be P <0.001 [T = 5.89, df = (1, 58), FWE-corrected] for each voxel with a cluster size of at least 40 voxels. Phip, parahippocampus;BA, Brodmann’s area; Hip, hippocampus; MTG,middle temporal gyrus; L, left; R, right;STG,superior temporal gyrus.

REFERENCES

[1] Ledberg A, Akerman S, Roland PE. Estimation of the probabilities of 3D clusters in functional brain images. Neuroimage 1998, 8: 113-128.

1/6