ChP segmentation accuracy
The manual annotation consistency was 0.800 among the five radiologists, compared to that was 0.608 among the five trainees. The proposed pipeline outperformed the FreeSurfer and the FastSurfer methods by 27.8% and 24.9% higher Dice coefficient, 24.8% and 22.6% higher Jaccard coefficient, 32.7% and 31.5% higher recall, and 66.3% and 53.5% lower Hausdorff distance, Table 2. The FreeSurfer underestimated the inferior horn of LVEN, while FastSurfer wrongly recognized some part of the gray matter as LVEN, Figure 2a. Both the FreeSurfer and FastSurfer underestimated the ChP, Figure 2b. In contrary, the proposed pipeline achieved more consistent segmentation as the overlapped regions of the radiologists.
Table 2. ChP segmentation performance quantitative comparison on the ChP multi-rater test dataset.
Metrics
|
Dice coefficient
|
Jaccard coefficient
|
Recall
|
Hausdorff distance
|
Proposed
|
0.620
|
0.458
|
0.721
|
14.817
|
FreeSurfer
|
0.342
|
0.210
|
0.394
|
44.041
|
FastSurfer
|
0.371
|
0.232
|
0.406
|
31.859
|
ChP Volumes in AD
Figure 3 shows the distributions of ChP and lateral ventricle volumes in each patient group. The Shapiro-Wilk test and Levene test showed that ChP volume in each group followed normal distributions and had homogeneous variances (p>0.05, Table S1 and S2). One-way ANOVA indicated that the mean ChP volumes among the five groups were significantly different (F=6.973, p<0.001). Specifically, the ChP volumes in the AD and LMCI groups are significantly greater than in the CN group (Tukey Post Hoc Test, p<0.05, Figure 3a and Table S3).
The LVEN volume did not satisfy the normality and variance homogeneity tests (Tables S1 and S2). Non-parametric independent-samples Kruskal-Wallis test showed significant differences in LVEN volumes among the five groups (H=53.351, p<0.0001). Pairwise comparisons of groups revealed significant differences in the distribution of LVEN volume between SMC-AD, CN-LMCI, CN-AD, EMCI-AD, and LMCI-AD, Figure 3b and Table S4.
ChP Blood Flow in AD
The relative perfusion of ChP in the three groups showed homogeneous variance but the CN group did not follow normal distribution (refer to Table S5 and S6 in the supplementary material), Figure 4. Non-parametric Mann-Whitney U-test showed that AD patients had significantly higher ChP relative perfusion compared to the CN group (Z=682.0, p=0.032).
ChP Volumes and Protein Levels
The histograms of protein levels showed normal distribution in Aβ40 and skew distributions in Aβ42, t-tau, and p-tau, Figure S1. Therefore, an arithmetic log transformation was applied to these proteins to meet the assumptions of the linear regression. Although significant correlations between the ChP volume and each protein level was found with all the subjects using the simple linear regression model, the associations may be not significant in each group, Table 3, Figure S2, and S3. Specifically, the ChP volume shows significant correlations with the p-tau and t-tau levels in each patient group, while it is not significantly correlated with the Aβ42 and Aβ40 in the CN, MCI, and AD groups.
Table 3. Simple linear regression between the ChP volume and p-tau, t-tau, Aβ42, and Aβ40. β is the standardized coefficient in the regression model
|
p-tau
|
t-tau
|
Aβ42
|
Aβ40
|
|
β
|
p
|
β
|
p
|
β
|
p
|
β
|
p
|
All
|
-0.136
|
0.001
|
-0.151
|
0.001
|
-0.263
|
0.001
|
-0.196
|
0.001
|
CN
|
-0.218
|
0.006
|
-0.209
|
0.009
|
-0.143
|
0.076
|
-0.241
|
0.001
|
SMC
|
-0.305
|
0.003
|
-0.292
|
0.004
|
-0.252
|
0.014
|
|
|
EMCI
|
-0.166
|
0.006
|
-0.185
|
0.002
|
-0.223
|
0.001
|
-0.161
|
0.112
|
LMCI
|
-0.216
|
0.007
|
-0.226
|
0.005
|
-0.273
|
0.001
|
|
|
AD
|
-0.359
|
0.001
|
-0.376
|
0.001
|
-0.156
|
0.078
|
0.028
|
0.882
|
Considering the LVEN volume, age, gender, and brain parenchyma volume (Pare), the ChP volume exhibited significant correlations with Aβ42, p-tau, and t-tau (p is from 0.012 to 0.001) in all subjects in the multiple linear regression model, Table 4. More importantly, the ChP volume is correlated with the p-tau and t-tau in most patient groups (p is from 0.027 to 0.001, except for the LMCI group) and the correlation coefficients, , are generally reducing in their absolute values along with the AD progression. In contrast, the Aβ42 and Aβ40 levels are significantly correlated (p is from 0.017 to 0.001) with the lateral ventricle volume mostly, except for the Aβ42 in the CN, EMCI and AD groups.
By controlling the colinear and excluding insignificant variables, the stepwise regression models show similar results as the multiple linear regression model with improved significance levels, Table 5. The ChP related variables, including ChP volume and ChP/parenchyma ratio, showed significant correlations (p is from 0.001 to 0.021) with p-tau and t-tau in each patient group. In contrast, the lateral ventricle related variables show significant correlations (p is from 0.001 to 0.027) with the Aβ42 and Aβ40 levels, except for the Aβ40 in MCI group. As expected, the age is also a significant contributor in most groups.
Table 4. Multiple linear regression in each group on p-tau, t-tau, Aβ42, and Aβ40. β is the standardized coefficient in the regression model
|
|
p-tau
|
t-tau
|
Aβ42
|
Aβ40
|
|
Var
|
β
|
p
|
β
|
p
|
β
|
p
|
β
|
p
|
Total
|
ChP
|
-0.200
|
0.001
|
-0.199
|
0.001
|
-0.111
|
0.012
|
-0.100
|
0.095
|
|
LVEN
|
-0.048
|
0.271
|
-0.076
|
0.078
|
-0.260
|
0.001
|
-0.392
|
0.001
|
|
Age
|
0.197
|
0.001
|
0.208
|
0.001
|
0.034
|
0.379
|
0.326
|
0.001
|
|
Gender
|
0.058
|
0.163
|
0.052
|
0.216
|
-0.026
|
0.519
|
-0.026
|
0.687
|
|
Pare
|
-0.128
|
0.002
|
-0.135
|
0.001
|
0.132
|
0.001
|
0.108
|
0.061
|
CN
|
ChP
|
-0.327
|
0.001
|
-0.306
|
0.001
|
-0.065
|
0.531
|
-0.234
|
0.005
|
LVEN
|
-0.150
|
0.092
|
-0.155
|
0.091
|
-0.166
|
0.103
|
-0.343
|
0.001
|
Age
|
0.593
|
0.001
|
0.541
|
0.001
|
-0.110
|
0.266
|
0.361
|
0.001
|
Gender
|
-0.122
|
0.170
|
-0.089
|
0.328
|
0.136
|
0.180
|
0.030
|
0.727
|
Pare
|
0.281
|
0.002
|
0.223
|
0.015
|
-0.055
|
0.583
|
0.165
|
0.034
|
SMC
|
ChP
|
-0.260
|
0.016
|
-0.237
|
0.027
|
-0.179
|
0.111
|
|
|
LVEN
|
-0.274
|
0.013
|
-0.309
|
0.005
|
-0.273
|
0.017
|
|
|
Age
|
0.368
|
0.001
|
0.363
|
0.001
|
0.060
|
0.586
|
|
|
Gender
|
-0.057
|
0.624
|
-0.024
|
0.834
|
0.182
|
0.135
|
|
|
Pare
|
0.092
|
0.442
|
0.035
|
0.766
|
-0.018
|
0.884
|
|
|
EMCI
|
ChP
|
-0.242
|
0.001
|
-0.257
|
0.001
|
-0.122
|
0.122
|
-0.054
|
0.613
|
LVEN
|
-0.225
|
0.002
|
-0.225
|
0.002
|
-0.036
|
0.635
|
-0.394
|
0.001
|
Age
|
0.453
|
0.001
|
0.457
|
0.001
|
-0.181
|
0.012
|
0.286
|
0.011
|
Gender
|
0.056
|
0.419
|
0.045
|
0.512
|
-0.008
|
0.909
|
-0.092
|
0.481
|
Pare
|
0.118
|
0.085
|
0.121
|
0.076
|
-0.106
|
0.143
|
0.090
|
0.452
|
LMCI
|
ChP
|
-0.239
|
0.046
|
-0.209
|
0.078
|
0.061
|
0.587
|
|
|
LVEN
|
-0.085
|
0.467
|
-0.135
|
0.248
|
-0.369
|
0.001
|
|
|
Age
|
0.121
|
0.204
|
0.151
|
0.111
|
0.084
|
0.345
|
|
|
Gender
|
0.031
|
0.757
|
-0.005
|
0.962
|
-0.181
|
0.057
|
|
|
Pare
|
-0.151
|
0.116
|
-0.125
|
0.190
|
0.307
|
0.001
|
|
|
AD
|
ChP
|
-0.209
|
0.037
|
-0.212
|
0.031
|
-0.116
|
0.275
|
0.238
|
0.194
|
|
LVEN
|
-0.184
|
0.071
|
-0.212
|
0.034
|
-0.177
|
0.103
|
-0.462
|
0.017
|
|
Age
|
-0.021
|
0.810
|
0.009
|
0.921
|
0.121
|
0.203
|
0.589
|
0.003
|
|
Gender
|
-0.124
|
0.224
|
-0.118
|
0.238
|
0.097
|
0.372
|
0.225
|
0.259
|
|
Pare
|
-0.026
|
0.794
|
-0.070
|
0.476
|
-0.067
|
0.530
|
-0.335
|
0.078
|
Table 5. Stepwise regression in each group for p-tau, t-tau, Aβ42, and Aβ40.
|
|
p-tau
|
t-tau
|
Aβ42
|
Aβ40
|
|
Var
|
β
|
p
|
β
|
p
|
β
|
p
|
β
|
p
|
Total
|
ChP/Pare
|
-0.227
|
0.001
|
|
|
-0.159
|
0.001
|
-0.125
|
0.026
|
|
ChP/LVEN
|
|
|
|
|
|
|
|
|
|
ChP
|
|
|
-0.220
|
0.001
|
|
|
|
|
|
LVEN
|
|
|
|
|
-0.234
|
0.001
|
-0.384
|
0.001
|
|
Age
|
0.193
|
0.001
|
0.197
|
0.001
|
|
|
0.317
|
0.001
|
|
Gender
|
|
|
|
|
|
|
|
|
|
Pare
|
-0.204
|
0.001
|
-0.120
|
0.001
|
|
|
|
|
CN
|
ChP/Pare
|
-0.487
|
0.001
|
-0.447
|
0.001
|
|
|
-0.248
|
0.001
|
ChP/LVEN
|
|
|
|
|
|
|
|
|
ChP
|
|
|
|
|
|
|
|
|
LVEN
|
|
|
|
|
-0.205
|
0.010
|
-0.308
|
0.001
|
Age
|
0.539
|
0.001
|
0.499
|
0.001
|
|
|
0.357
|
0.001
|
Gender
|
|
|
|
|
|
|
|
|
Pare
|
|
|
|
|
|
|
|
|
SMC
|
ChP/Pare
|
|
|
|
|
|
|
|
|
ChP/LVEN
|
|
|
|
|
|
|
|
|
ChP
|
-0.270
|
0.010
|
-0.242
|
0.021
|
|
|
|
|
LVEN
|
-0.262
|
0.015
|
-0.304
|
0.005
|
-0.325
|
0.001
|
|
|
Age
|
0.341
|
0.001
|
0.353
|
0.001
|
|
|
|
|
Gender
|
|
|
|
|
|
|
|
|
Pare
|
|
|
|
|
|
|
|
|
EMCI
|
ChP/Pare
|
-0.256
|
0.001
|
-0.269
|
0.001
|
|
|
|
|
ChP/LVEN
|
|
|
|
|
|
|
|
|
ChP
|
|
|
|
|
-0.167
|
0.009
|
|
|
LVEN
|
-0.208
|
0.002
|
-0.213
|
0.001
|
|
|
-0.424
|
0.001
|
Age
|
0.446
|
0.001
|
0.450
|
0.001
|
-0.144
|
0.025
|
0.233
|
0.016
|
Gender
|
|
|
|
|
|
|
|
|
Pare
|
|
|
|
|
|
|
|
|
LMCI
|
ChP/Pare
|
|
|
|
|
|
|
|
|
ChP/LVEN
|
|
|
|
|
|
|
|
|
ChP
|
-0.237
|
0.003
|
-0.247
|
0.002
|
|
|
|
|
LVEN
|
|
|
|
|
-0.357
|
0.001
|
|
|
Age
|
|
|
|
|
|
|
|
|
Gender
|
|
|
|
|
|
|
|
|
Pare
|
-0.174
|
0.028
|
-0.176
|
0.026
|
0.202
|
0.007
|
|
|
AD
|
ChP/Pare
|
|
|
|
|
|
|
|
|
|
ChP/LVEN
|
|
|
|
|
|
|
0.547
|
0.004
|
|
ChP
|
-0.246
|
0.011
|
-0.245
|
0.010
|
|
|
|
|
|
LVEN
|
-0.220
|
0.022
|
-0.255
|
0.007
|
-0.195
|
0.027
|
|
|
|
Age
|
|
|
|
|
|
|
0.629
|
0.001
|
|
Gender
|
|
|
|
|
|
|
|
|
|
Pare
|
|
|
|
|
|
|
|
|