| 21 | 0 | 52 |
| 下载次数 | 被引频次 | 阅读次数 |
基于多模态磁共振成像的图神经网络因能有效表征脑网络的非欧几里得结构，在医学影像领域备受关注。然而，现有研究大多忽略了患者症状的显著个体差异，仅通过简单拼接引入人口统计学数据，限制了分类性能。此外，多站点数据的异质性也导致模型在跨站点外部验证中表现不佳。为此提出了一种基于人口统计学调控的图注意力神经网络框架（demographic-regulated multimodal graph attention network framework，DM-GAT）。首先设计人口感知动态构图模块（demographic-aware dynamic graph structure learning，D-GSL），利用人口统计学信息重构潜在的脑网络拓扑结构，解决静态功能连接中的噪声与个体差异问题。其次，提出多视角人口交互注意力机制（multi-view demographic interaction attention，M-DIA），通过引入人口学特征作为门控信号，动态调控节点与边的注意力权重，增强个体化表征能力。在ABIDE和ADHD-200跨站点数据集上的实验结果表明，提出的模型在ABIDE数据集（17个站点，943名受试者）上的分类准确率（Accuracy， ACC）达到74.51%，优于先进模型MHNet（73.27%）。消融实验进一步证实了D-GSL模块（ACC提升约4%）、M-DIA模块的有效性。
Abstract：Graph Neural Networks (GNNs) based on multimodal Magnetic Resonance Imaging (MRI) have garnered significant attention in medical imaging due to their ability to effectively characterize the non-Euclidean structure of brain networks. However， most existing studies overlook significant individual heterogeneity in patient symptoms and merely introduce demographic data via simple concatenation， thereby limiting classification performance. Furthermore， the heterogeneity of multi-site data often results in suboptimal model performance during cross-site external validation. To address these challenges， this study proposes a demographic-regulated multimodal graph attention network framework (DM-GAT). First， the model incorporates a demographic-aware dynamic graph structure learning (D-GSL) module， which utilizes demographic information to reconstruct the latent topology of brain networks， thereby mitigating noise and individual variability issues inherent in static functional connectivity. Second， a multi-view demographic interaction attention (M-DIA) mechanism is introduced. This mechanism employs demographic features as gating signals to dynamically regulate the attention weights of nodes and edges， enhancing the model's capability for individualized representation. Experimental results on the ABIDE and ADHD-200 cross-site datasets demonstrate that the proposed model achieves a classification accuracy of 74.51% on the ABIDE dataset (17 sites， 943 subjects)， outperforming the state-of-the-art MHNet model (73.27%). Ablation studies further validate the effectiveness of the D-GSL module (improving accuracy by approximately 4%) and the M-DIA module.
[1] 翁旭初，静进，罗建红，等，孤独症谱系障碍脑影像辅助诊断与经颅磁刺激治疗进展[J].中国心理卫生杂志，2025， 39(8):661-670. WENG X C， JING J， LUO J H， et al. Neuroimaging aided diagnosis and transcranial magnetic stimulation interventions for autism spectrum disorder[J]. Chinese Mental Health Journal， 2025， 39(8): 661-670. (in Chinese).
[2] LYNCH F L， BULKLEY J E， VARGA A， et al. The impact of autism spectrum disorder on parent employment: Results from the r-Kids study[J]. Autism Research， 2023， 16(3): 642-652..
[3] 肖石花，李晶，自然发展行为干预在孤独症早期干预中的应用[J].心理科学进展，2023，31(12): 2350-2367. XIAO S H， LI J. Implementation of Naturalistic Developmental Behavioral Interventions: an early intervention program for children with autism spectrum disorder[J]. Advances in Psychological Science， 2023， 31(12): 2350-2367. (in Chinese).
[4] PIZZOLORUSSO F， PAPARELLA M T， PIZZOLORUSSO I， et al. Magnetic Resonance Imaging in Autism Spectrum Disorders: clinical and neuroradiological phenotypes[J]. Acta Bio-Medica， 2023， 94(2): e2023027.
[5] KABASAWA H. MR imaging in the 21st century: technical innovation over the first two decades[J]. Magnetic Resonance in Medical Sciences， 2022， 21(1): 71-82.
[6] BISWAL B B， UDDIN L Q. The history and future of resting-state functional magnetic resonance imaging[J]. Nature， 2025， 641(8065): 1121-1131.
[7] BELHAOUARI S B， TALBI A， HASSAN S， et al. PFT: a novel time-frequency decomposition of BOLD fMRI signals for autism spectrum disorder detection[J]. Sustainability， 2023， 15(5): 4094.
[8] TANG C， WEI Y Q， ZHAO J J， et al. The Dynamic Measurements of Regional Brain Activity for Resting-State fMRI: d-ALFF， d-fALFF and d-ReHo[C]//Medical Image Computing and Computer Assisted Intervention MICCAI 2018. Cham: Springer， 2018: 190-197.
[9] 何苑敏，杨光钊，茅国群，等，注意缺陷多动障碍患儿功能磁共振低频振幅与持续性注意力相关性研究[J].浙江医学，2025， 47(11): 1180-1183. HEY M. YANG G Z， MAO G Q， et al. Application of resting-state fMRI low frequency fluctuation for evaluation of sustained attention in children with attention deficit hyperactivity disorder[J]. Zhejiang Medical Journal， 2025， 47(11):1180-1183. (in Chinese)
[10] ZHANG J， FENG F， HAN T Y， et al. Detection of autism spectrum disorder using fMRI functional connectivity with feature selection and deep learning[J]. Cognitive Computation， 2023， 15(4): 1106-1117.
[11] YüCELBA? ?， YüCELBA? C. Autism spectrum disorder detection using sequential minimal optimization-support vector machine hybrid classifier according to history of jaundice and family autism in children[J]. Concurrency and Computation: Practice and Experience， 2022， 34: e6498.
[12] LAMANI MR， BENADIT P J， VAITHINATHAN K. Multi-atlas graph convolutional networks and convolutional recurrent neural networks-based ensemble learning for classification of autism spectrum disorders [J]. SN Computer Science， 2023， 4(3): 213.
[13] VENKATA RATNA PRABHA K， BINDU C H， DEVI K R. Optimized attention-enhanced U-Net for autism detection and region localization in MRI[J]. Psychiatry Research: Neuroimaging， 2025， 349: 111970.
[14] TANG J， CHEN J， HU M J， et al. Diagnosis of autism spectrum disorder (ASD) by dynamic functional connectivity using GNN-LSTM[J]. Sensors， 2025， 25(1): 156.
[15] YANG C D， WANG P Y， TAN J， et al. Autism spectrum disorder diagnosis using graph attention network based on spatial-constrained sparse functional brain networks[J]. Computers in Biology and Medicine， 2021， 139: 104963.
[16] XIAO S X， WANG S P， DAI Y F， et al. Graph neural networks in node classification: survey and evaluation[J]. Machine Vision and Applications， 2022， 33: 4.
[17] 王海鹏，高自强，董佳俊，等，基于粗糙注意力融合机制与 Group Transformer 的视网膜血管分割网络[J].南通大学学报(自然科学版)，2024，23(1): 28-37， 48. WANG H P， GAO Z Q， DONG J J， et al. Retinal vessel segmentation network based on rough attention fusion mechanism and Group Transformer[J]. Journal of Nantong University (Natural Science Edition)， 2024， 23(1): 28-37， 48. (in Chinese)
[18] 高轩瑞，曹进德，张金壬，一种基于深度学习的心电图分类方法[J].南通大学学报(自然科学版)，2023，22(3):17-25. GAO X R， CAO J D， ZHANG J R. A method of ECG classification based on deep learning[J]. Journal of Nantong University (Natural Science Edition)， 2023， 22(3): 17-25. (in Chinese)
[19] YE Y C， LI Y H， OUYANG R L， et al. Improving machine learning based phase and hardness prediction of high-entropy alloys by using Gaussian noise augmented data[J]. Computational Materials Science， 2023， 223: 112140.
[20] Orvieto A， Kersting H， Proske F， et al. Anticorrelated noise injection for improved generalization [C] //International Conference on Machine Learning. PMLR， 2022: 17094-17116.
[21] DI MARTINO A， YAN C G， LI Q， et al. The autism brain imaging data exchange: towards a large-scale evaluation of the intrinsic brain architecture in autism[J]. Molecular Psychiatry， 2014， 19(6): 659-667.
[22] CONSORTIUM T. The ADHD-200 Consortium: a model to advance the translational potential of neuroimaging in clinical neuroscience[J]. Frontiers in Systems Neuroscience， 2012， 6: 62.
[23] WAGHELA H， SEN J， RAKSHIT S. Robust image classification: defensive strategies against FGSM and PGD adversarial attacks [C]//Proceedings of the 2024 Asian Conference on Intelligent Technologies (ACOIT)， September 6-7， 2024， KOLAR， India. New York: IEEE， 2024: 1-7.
[24] LI GH， MULLER M， THABET A， et al. DeepGCNs: can GCNs go as deep as CNNs? [C]//Proceedings of the 2019 IEEE/CVF International Conference on Computer Vision (ICCV)， October 27-November 2， 2019. Seoul， Korea. New York: IEEE， 2019: 9266-9275.
[25] LIX X， ZHOU Y， DVORNEK N， et al. BrainGNN: interpretable brain graph neural network for fMRI analysis[J]. Medical Image Analysis， 2021， 74: 102233.
[26] CHEN Y Z， YAN J D， JIANG M X， et al. Adversarial learning based node-edge graph attention networks for autism spectrum disorder identification[J]. IEEE Transactions on Neural Networks and Learning Systems， 2024， 35(6): 7275-7286.
[27] WEN G Q， CAO P， BAO H W， et al. MVS-GCN: a prior brain structure learning-guided multi-view graph convolution network for autism spectrum disorder diagnosis[J]. Computers in Biology and Medicine， 2022， 142: 105239.
[28] JANG H， CAO P， XU M Y， et al. Hi-GCN: a hierarchical graph convolution network for graph embedding learning of brain network and brain disorders prediction[J]. Computers in Biology and Medicine， 2020， 127: 104096.
[29] KAWAHARA J， BROWN C J， MILLER S P， et al. BrainNetCNN: Convolutional neural networks for brain networks; towards predicting neurodevelopment[J]. NeuroImage， 2017， 146: 1038-1049.
[30] CUI H J， DAI W， GUO Y， et al. Brain network transformer[C]//Proceedings of the Advances in Neural Information Processing Systems 35， November 28-December 9， 2022. New Orleans， Louisiana， USA. Neural Information Processing Systems Foundation， Inc. (NeurIPS)， 2022: 25586-25599.
[31] LIU M X， ZHANG H， SHI F， et al. Hierarchical graph convolutional network built by multiscale atlases for brain disorder diagnosis using functional connectivity [J]. IEEE Transactions on Neural Networks and Learning Systems， 2024， 35(11): 15182-15194.
[32] BANNADABHAVI A， LEE S， DENG W L， et al. Community-aware transformer forAutism prediction infMRI connectome[C]//Medical Image Computing and Computer Assisted Intervention MICCAI 2023. Cham: Springer， 2023: 287-297.
[33] LI Y Y， ZENG W M， DONG W H， et al. MHNet: multi-view high-order network for diagnosing neurodevelopmental disorders using resting-state fMRI[J]. Journal of Imaging Informatics in Medicine， 2025， 38(5): 2994-3014.
[34] Yu S， Jin S， Li M， et al. Long-range brain graph transformer[J]. Advances in Neural Information Processing Systems， 2024， 37: 24472-24495.
[35] Wei X， Zhao K， Jiao Y， et al. Multi-modal cross-domain self-supervised pre-training for fMRI and EEG fusion[J]. Neural Networks， 2025， 184: 107066.
基本信息:
中图分类号:TP183;R749.94
引用信息:
[1]张雄涛,麻远,王丰,等.一种基于人口调控注意力网络的孤独症辅助诊断模型[J].南通大学学报(自然科学版)().
2026-06-26
2026-06-26
2026-06-26