Graph neural networks have achieved remarkable progress in the early diagnosis of Alzheimer's disease (AD). However, most existing methods suffer from low-quality graph construction, making it difficult to accurately model the topological relationships between nodes under noisy samples condition, which in turn affects classification performance. To address this issue, a classification network named GFGCN-SP is proposed. Instead of relying on the single graph construction strategy, GFGCN-SP employs the multi-graph fusion approach to enhance the quality of the graph structure. Additionally, a contrastive learning mechanism is introduced to guide graph pooling, enabling effective hierarchical representation learning. Extensive experiments conducted on the ADNI dataset across three classification tasks (AD-CN, MCI-CN, and EMCI-LMCI) demonstrate that the proposed method significantly outperforms existing state-of-the-art baselines, achieving average classification accuracies of 92.64%, 86.85%, and 79.04%, respectively, thereby validating its effectiveness in early-stage AD identification.
静息态功能磁共振成像(resting-state functional Magnetic Resonance Imaging,rs-fMRI)是一种非入侵性成像技术,可以捕获大脑的结构变化,因此已经广泛用于AD诊断领域.在利用rs-fMRI进行脑网络分析时,将脑建模为由一组节点(例如感兴趣的脑区(Region of Interest,ROI)[4])和衡量节点间功能连接的连接矩阵[5]组成的网络,以此来增加对人脑功能结构的理解.脑网络节点则根据解剖脑谱图、脑功能分区或者脑功能分解来定义,这些节点通过其局部形状、功能一致性或局部连接度量来表征[6],脑网络节点之间的功能连接通过其rs-fMRI时间序列间的时间相关性来衡量.如何使用rs-fMRI进行准确诊断仍是当前研究的重要挑战[7].
SIMONR P, AMINOFFM J, GREENBERGD A. Clinical neurology[M]. New York: Lange Medical Books/McGraw-Hill, 2009.
[2]
SHARMAP, SRIVASTAVAP, SETHA, et al. Comprehensive review of mechanisms of pathogenesis involved in Alzheimer’s disease and potential therapeutic strategies[J]. Progress in Neurobiology, 2019, 174: 53-89.
[3]
SCHELTENSP, DE STROOPERB, KIVIPELTOM, et al. Alzheimer’s disease[J]. The Lancet, 2021, 397(10284): 1577-1590.
[4]
ZHOUZ, WANGQ, ANX, et al. A novel graph neural network method for Alzheimer’s disease classification[J]. Computers in Biology and Medicine, 2024, 180: 108869.
[5]
DECOG, KRINGELBACHM L. Great expectations: Using whole-brain computational connectomics for understanding neuropsychiatric disorders[J]. Neuron, 2014, 84(5): 892-905.
[6]
BLUMENSATHT, JBABDIS, GLASSERM F, et al. Spatially constrained hierarchical parcellation of the brain with resting-state fMRI[J]. NeuroImage, 2013, 76: 313-324.
[7]
SMITHS M, MILLERK L, SALIMI-KHORSHIDIG, et al. Network modelling methods for FMRI[J]. NeuroImage, 2011, 54(2): 875-891.
[8]
OONOK, SUZUKIT. Graph neural networks exponentially lose expressive power for node classification[C]// 8th International Conference on Learning Representations.Addis Ababa:ICLR, 2020: 26-30.
[9]
LIX, ZHOUY, DVORNEKN, et al. BrainGNN: Interpretable brain graph neural network for fMRI analysis[J]. Medical Image Analysis, 2021, 74: 102233.
[10]
YAOD, SUIJ, YANGE, et al. Temporal-adaptive graph convolutional network for automated identification of major depressive disorder using resting-state fMRI[C]// Machine Learning in Medical Imaging. Cham: Springer, 2020: 1-10.
[11]
GANJ, PENGZ, ZHUX, et al. Brain functional connectivity analysis based on multi-graph fusion[J]. Medical Image Analysis, 2021, 71: 102057.
[12]
LEEJ, LEEI, KANGJ. Self-attention graph pooling[C]//International Conference on Machine Learning. Long Beach: PMLR, 2019: 3734-3743.
[13]
RANJANE, SANYALS, TALUKDARP. ASAP: Adaptive structure aware pooling for learning hierarchical graph representations[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(4): 5470-5477.
[14]
LIZ, TANGC, LIUX, et al. Consensus graph learning for multi-view clustering[J]. IEEE Transactions on Multimedia, 2021, 24: 2461-2472.
[15]
ZHAOQ, ZHANGH, HEM, et al. Graph pooling via dual-view multi-level infomax[J]. Knowledge-Based Systems, 2023, 260: 110089.
[16]
XIEY, YAOC, GONGM, et al. Graph convolutional networks with multi-level coarsening for graph classification[J]. Knowledge-Based Systems, 2020, 194: 105578.
[17]
YANC G, WANGX D, ZUOX N, et al. DPABI: Data processing & analysis for (resting-state) brain imaging[J]. Neuroinformatics, 2016, 14(3): 339-351.
[18]
DVORNEKN C, YANGD, VENTOLAP, et al. Learning generalizable recurrent neural networks from small task-fMRI datasets[C]// Medical Image Computing and Computer Assisted Intervention. Cham: Springer, 2018: 329-337.