Organizations: School of Information and Electronic Engineering, Zhejiang Gongshang University, Hangzhou 310018, China · School of Computer and Network Security, Chengdu University of Technology, Chengdu, 610059, China · School of Computing Technologies, RMIT University, Melbourne, VIC 3000, Australia
Neurodegenerative disorders are characterized by their progressive nature, with distinct stages that require precise identification and tracking of progression trajectories for accurate diagnosis. However, existing diagnostic methods often fail to fully capture the unique brain patterns associated with different stages, limiting their ability to effectively target specific stages of the disorders. To address this issue, we propose a Stage-aware Multimodal Brain Graph Contrastive Learning (smBGC) method. Specifically, smBGC introduces a stage-aware feature enhancement strategy to capture stage-specific features across all progression stages, enabling the generation of stage-aware multimodal brain graph representations. In addition, smBGC incorporates a stage-adaptive brain graph augmentation approach within a graph contrastive learning framework, facilitating multimodal fusion and improving the model's capability to distinguish between different disorder stages. Experiments on two real-world datasets demonstrate the superior performance of smBGC in various neurodegenerative disorder diagnosis tasks compared with state-of-the-art methods.
Figures & tables
Notations
Description
B
Dataset of brain graphs
Gis , Gif , Gi∗
SBG, FBG and a brain graph from any modality of the i-th subject, respectively
yi
The stage label of i-th subject
V
Node set of i-th subject
Ei∗
Edge set of Gi∗ , ∗ represents any modality
M
The number of subjects in B
TABLE I : Notations in this paper.
Fig. 1 : The overall framework of smBGC. Multimodal data are used to construct the FBG and SBG. The feature score matrix of each stage is generated using SVD, and feature enhancement is performed with weighted vectors. Both structural and functional brain graphs undergo feature masking and enhancement, with the new graph representations input into a parameter-sharing GNN for contrastive learning to capture the associations and differences between FBGs and SBGs.
Fig. 2 : Overview of the proposed Communal Feature Dropping.
Fig. 3 : Overview of the proposed stage-adaptive brain graph augmentation approach. We select the brain graph of a single subject from one stage as an example to illustrate how smBGC utilizes the scoring matrix for graph augmentation.
Hyperparameter
ADNI-3 (three stages)
ADNI-3 (four stages)
PPMI
Functional magnitude αf
0.6
0.6
0.7
Structural magnitude αs
0.6
0.5
0.3
Dropping features λ
20
20
30
GNN layers
1
1
1
Dropout
0.5
0.5
0.5
Readout method
sum
sum
sum
TABLE II : Hyperparameters for training on three different tasks.
Method
NC vs. MCI vs. AD
NC vs. EMCI vs. LMCI vs. AD
ACC
F1
AUC
SEN
SPE
ACC
F1
AUC
SEN
SPE
GCN
54.30 ± 1.67
41.97 ± 2.29
67.51 ± 2.72
42.80 ± 1.79
74.06 ± 1.19
43.75 ± 3.64
35.01 ± 3.60
65.90 ± 3.58
35.07 ± 3.60
80.83 ± 1.29
GAT
50.78 ± 2.44
41.65 ± 3.48
65.22 ± 2.99
41.79 ± 2.86
72.07 ± 0.88
46.88 ± 7.23
37.07 ± 5.31
69.95 ± 5.03
39.17 ± 4.77
81.38 ± 1.87
GIN
50.77 ± 2.57
42.38 ± 2.48
63.06 ± 2.45
43.19 ± 2.75
72.89 ± 1.42
50.00 ± 11.63
47.28 ± 11.58
63.51 ± 7.34
47.22 ± 10.95
82.56 ± 3.93
GraphSAGE
56.95 ± 4.13
47.74 ± 5.15
69.47 ± 3.56
47.84 ± 5.02
75.55 ± 2.87
43.75 ± 6.12
39.29 ± 5.14
61.55 ± 4.87
40.84 ± 4.46
81.26 ± 2.47
BrainGB
53.33 ± 4.46
38.23 ± 4.79
62.10 ± 3.09
40.84 ± 4.12
72.35 ± 2.81
40.63 ± 5.92
32.14 ± 6.02
55.68 ± 4.01
34.67 ± 4.20
76.02 ± 1.79
TABLE III : Performance comparison of different methods on the ADNI-3 dataset (%). The best results are highlighted in bold and the second-best results are underlined.
Method
NC vs. Prodromal vs. PD
ACC
F1
AUC
SEN
SPE
GCN
51.92 ± 3.57
49.55 ± 3.19
66.26 ± 4.91
50.34 ± 2.23
74.66 ± 0.66
GAT
50.58 ± 5.66
46.22 ± 1.68
60.32 ± 3.08
48.00 ± 1.62
73.20 ± 5.56
GIN
50.52 ± 5.10
49.24 ± 4.23
68.13 ± 3.94
49.83 ± 4.42
74.25 ± 2.43
GraphSAGE
55.04 ± 2.67
52.46 ± 5.20
66.91 ± 5.77
53.22 ± 4.35
76.19 ± 1.29
BrainGB
48.10 ± 5.39
42.80 ± 4.26
60.82 ± 2.62
45.51 ± 2.99
72.29 ± 1.75
TABLE IV : Performance comparison of different methods on the PPMI dataset (%).
Metric
Fold 1
Fold 2
Fold 3
Fold 4
Fold 5
Mean ± Std.
95% CI
ACC
80.12
82.68
82.90
82.11
80.74
81.71±1.22
[80.19, 83.23]
F1
78.43
79.12
79.78
79.25
78.52
79.02±0.56
[78.33, 79.71]
AUC
96.55
96.28
95.73
96.51
95.63
96.14±0.43
[95.60, 96.68]
SEN
62.42
67.21
63.85
67.54
63.83
64.97±2.27
[62.15, 67.79]
SPE
96.72
95.81
96.65
95.54
96.48
96.24±0.53
[95.58, 96.90]
TABLE V : Fold-level performance of smBGC on the four-stage classification task of the ADNI-3 dataset (%).
Metric
smBGC (%)
MHGF (%)
Improvement
95% CI
t(4)
Adjusted p
ACC
81.71
57.56
24.15
[20.71, 27.59]
19.47
<0.001
F1
79.02
47.41
31.61
[27.84, 35.37]
23.29
<0.001
AUC
96.14
69.98
26.16
[19.65, 32.67]
11.16
<0.001
SEN
64.97
50.78
14.19
[11.50, 16.87]
14.65
<0.001
SPE
96.24
85.23
11.01
[9.75, 12.28]
24.22
<0.001
TABLE VI : Two-sided paired t-tests results comparing smBGC with MHGF on the ADNI-3 four-stage classification task.
Fig. 4 : The t-SNE visualizations of different methods on the ADNI-3 dataset with four stages.
Fig. 5 : Experimental results under different modal choices.
Methods
NC vs. Prodromal vs. PD
ACC
F1
AUC
SEN
SPE
w/o CFD & ASFE
41.73 ± 5.26
36.97 ± 5.87
56.50 ± 4.08
39.53 ± 3.85
69.09 ± 5.43
w/o ASFE
57.44 ± 5.29
50.15 ± 1.59
70.94 ± 6.76
52.24 ± 5.63
80.88 ± 2.94
w/o CFD
83.24 ± 2.29
78.28 ± 3.61
97.47 ± 1.19
82.96 ± 2.64
91.73 ± 5.91
smBGC (ours)
91.15 ± 3.19
91.24 ± 3.13
96.27 ± 0.91
91.77 ± 4.33
95.61 ± 3.52
TABLE VII : Results of smBGC variants w/o each proposed component on the PPMI dataset.
Fig. 6 : The classification accuracy of smBGC with varied different hyperparameters on (a) ADNI-3 with three stages and (b) PPMI datasets.
Fig. 7 : Accuracy, F1-score, and AUC values of the proposed model under different neuron numbers over three datasets.
Department of Computer Science, Kennesaw State University, Marietta, GA, 30060 · Department of Information Technology, Kennesaw State University, Marietta, GA, 30060 · School of Data Science and Analytics, Kennesaw State University, Marietta, GA, 30060 +1