| 13 | 0 | 31 |
| 下载次数 | 被引频次 | 阅读次数 |
阿尔茨海默病（Alzheimer’s disease,AD）作为进行性神经退行性疾病，其病理机制涉及灰质萎缩、白质纤维破坏与全脑功能连接异常的复杂交互，传统单模态影像诊断模型因难以整合多维度病理信息而存在性能瓶颈.针对这一问题，提出一种结构核磁共振成像组织分解的多组织融合增强神经网络架构——DeMoNet，实现AD高精度分类与模型可解释性的统一.首先，该架构以结构磁共振成像组织分割得到灰质和白质3D影像为双分支并行输入，保留两类组织的三维空间拓扑信息与差异化病理特征.其次，双分支特征经包含3D卷积、双层归一化与ReLU激活的下采样模块完成多尺度特征初步编码与降维.双路径输出通过元素级相乘的门控机制完成特征融合，以全局关联特征引导局部病理特征权重分配，实现二者精准耦合.最后，融合特征经卷积与全连接层映射，输出AD与正常对照的二分类预测结果 .该架构的核心创新在于多组织输入、双路径特征提取与注意力增强的深度结合，既解决了传统卷积模型全局依赖不足与3D多头自注意力模型局部细节丢失的问题，又通过自适应特征筛选提升诊断精度.实验结果表明，AD与正常对照的分类精度约为97.20%，该架构不仅实现了AD的高精度诊断，同时为临床病理定位提供直观参考，推动人工智能诊断模型从“黑箱”向可解释方向跨越，具有重要的临床转化价值.
Abstract：Alzheimer's disease(AD), as a progressive neurodegenerative disorder, involves complex interactions among gray matter(GM) atrophy, white matter(WM) fiber disruption, and aberrant whole-brain functional connectivity. Conventional single-modality imaging diagnostic models face performance limitations due to their difficulty in integrating multidimensional pathological information. To address this issue, this paper proposed DeMoNet, a multi-tissue fusion-enhanced neural network architecture based on structural magnetic resonance imaging(sMRI) tissue decomposition, which achieved both highprecision AD classification and model interpretability. First, the architecture utilized 3D images of gray matter and white matter obtained from tissue segmentation of sMRI as dual-branch parallel inputs, preserving the three-dimensional spatial topological information and distinct pathological characteristics of both tissue types. Subsequently, the dual-branch features were processed through down-sampling modules comprising 3D convolution, dual normalization layers, and ReLU activation for preliminary multi-scale feature encoding and dimensionality reduction. The outputs of the two paths were fused through an element-wise multiplication-based gating mechanism, where global associative features guided the weighting of local pathological features, achieving precise coupling between them. Finally, the fused features were mapped through convolutional and fully connected layers to output binary classification predictions of AD versus normal controls. The core innovation of this architecture lay in the deep integration of multi-tissue input, dual-path feature extraction, and attention enhancement, which not only addressed the limitations of insufficient global dependency modeling in traditional convolutional models and the loss of local details in 3D multi-head self-attention models but also improved diagnostic accuracy through adaptive feature selection. Experimental results demonstrated that the classification accuracy for AD versus normal controls was approximately 97.20%. The proposed architecture not only achieved high-precision AD diagnosis but also provided intuitive references for clinical pathological localization, advancing AI diagnostic models from “black-box” systems to interpretable systems and offering significant potential for clinical translation.
[1]DUBOIS B,VILLAIN N,FRISONI G B,et al. Clinical diagnosis of Alzheimer’s disease:Recommendations of the International Working Group[J].The Lancet Neurology,2021,20(6):484-496.
[2]áVILA-VILLANUEVA M,MARCOS DOLADO A,GóMEZ-RAMíREZ J,et al.Brain structural and functional changes in cognitive impairment due to Alzheimer’s disease[J].Frontiers in Psychology,2022,13:886619.
[3]LIU Y M,CHEN S Y,LI S,et al.Diagnostic and inclusion criteria in Alzheimer’s disease clinical trials:A systematic review of the past decade[J]. Journal of Alzheimer’s Disease Reports,2025,9:25424823251362444.
[4]KAMNITSAS K,LEDIG C,NEWCOMBE V F J,et al.Efficient multi-scale3D CNN with fully connected CRF for accurate brain lesion segmentation[J].Medical Image Analysis,2017,36:61-78.
[5]JACK C R Jr,ARANI A,BOROWSKI B J,et al.Overview of adni mri[J].Alzheimer’s&Dementia,2024,20(10):7350-7360.
[6]KOOI T,LITJENS G,VAN GINNEKEN B,et al. Large scale deep learning for computer aided detection of mammographic lesions[J].Medical Image Analysis,2017,35:303-312.
[7]VOGEL J W,CORRIVEAU-LECAVALIER N,FRANZMEIER N,et al.Connectome-based modelling of neurodegenerative diseases:Towards precision medicine and mechanistic insight[J].Nature Reviews Neuroscience,2023,24(10):620-639.
[8]ZHAO Y M,JIANG B,CHEN Y,et al.Multi-modal orthogonal fusion network via cross-layer guidance for Alzheimer’s disease diagnosis[J].Neural Networks,2026,194:108091.
[9]BAI J,ZHANG Z Y,YIN Y,et al.LGG-NeXt:A next generation CNN and transformer hybrid model for the diagnosis of Alzheimer’s disease using 2D structural MRI[J]. IEEE Journal of Biomedical and Health Informatics,2025,29(4):2808-2818.
[10]ZUBAIR M,JAFFAR A,HUSSAIN S,et al.Alzheimer’s disease classification using a hybrid deep learning approach with multi-layer U-Net segmentation and XAI driven analysis[J]. PLoS One,2025,20(9):e0332572.
[11]LI L L,YU X F,SHENG C,et al.A review of brain imaging biomarker genomics in Alzheimer’s disease:Implementation and perspectives[J].Translational Neurodegeneration,2022,11:42.
[12]KIM T H,KIM B,KIM Y R,et al.Gray matter differences associated with menopausal hormone therapy in menopausal women:A DARTELbased VBM study[J].Scientific Reports,2023,13:1401.
[13]黄和生，陈文杰，戴舒霆，等.双分支多尺度网络的阿尔茨海默病诊断研究[J].激光与光电子学进展，2026,63(2):529-542.
[14]FENG J J,BA M W,LI N,et al.3D Multi-feature fusion convolutional network for Alzheimer’s disease diagnosis[C]//2024 46th Annual International Conference of the IEEE Engineering in Medicine and Biology Society(EMBC).July 15-19,2024.Orlando,FL,USA.IEEE,2024:1-6.
[15]LIU F L,REN X C,ZHANG Z Y,et al.Rethinking skip connection with layer normalization[C]//Proceedings of the 28th International Conference on Computational Linguistics. Barcelona,Spain(Online). International Committee on Computational Linguistics,2020:3586-3598.
[16]JANG J G,QUAN C,LEE H D,et al.Falcon:Lightweight and accurate convolution based on depthwise separable convolution[J].Knowledge and Information Systems,2023,65(5):2225-2249.
[17]WOO S,PARK J,LEE J Y,et al.CBAM:Convolutional block attention module[M]//Computer Vision-ECCV 2018.Cham:Springer,2018:3-19.
[18]ZHAO F,FENG F,YE S X,et al.Multi-head self-attention mechanismbased global feature learning model for ASD diagnosis[J]. Biomedical Signal Processing and Control,2024,91:106090.
[19]DAI Y L,ZOU B J,KUI X Y,et al.Strong multimodal representation learner through cross-domain distillation for Alzheimer’s disease classification[C]//2024 IEEE International Conference on Bioinformatics and Biomedicine(BIBM).December 3-6,2024.Lisbon,Portugal.IEEE,2024:3144-3149.
[20]MOHSIN N A,ABDUL AMEER M H. A dual stream deep learning framework for Alzheimer’s disease detection using MRI sonification[J].Journal of Imaging,2026,12(1):46.
[21]HE F Y,WU R Q,ZENG X Q,et al.Skin lesion classification network based on improved MobileViT[J].Engineering Applications of Artificial Intelligence,2025,159:111726.
[22]ZHONG S C,LOU J J,MA K,et al.Disentangling in-vivo microstructural changes of white and gray matter in mild cognitive impairment and Alzheimer’s disease:A systematic review and meta-analysis[J].Brain Imaging and Behavior,2023,17(6):764-777.
[23]DAI Y L,ZOU B J,ZHU C Z,et al. DE-JANet:A unified network based on dual encoder and joint attention for Alzheimer’s disease classification using multi-modal data[J]. Computers in Biology and Medicine,2023,165:107396.
[24]HOU T,ZHU H Q,LIU J H,et al.Integrating Gaussian mixture model with adjacent spatial adaptive transformer multi-stage network for magnetic resonance image reconstruction[J].Biomedical Signal Processing and Control,2025,104:107577.
[25]YUAN Z H,CHEN J Q,QIU H,et al. Co-activation pattern analysis based on hidden semi-Markov model for brain spatiotemporal dynamics[J].IEEE Transactions on Medical Imaging,2026,45(2):843-852.
[26]WAN D,CAI M C,PENG S F,et al. Deepfake detection algorithm based on dual-branch data augmentation and modified attention mechanism[J].Applied Sciences,2023,13(14):8313.
[27]XIN J M,WANG A C,GUO R,et al.CNN and swin-transformer based efficient model for Alzheimer’s disease diagnosis with sMRI[J].Biomedical Signal Processing and Control,2023,86:105189.
[28]QIANG Y R,ZHOU Q Y,LI J N,et al.Classification of Alzheimer’s disease by jointing 3D depthwise separable convolutional neural network and transformer[J].Expert Systems with Applications,2025,286:127720.
[29]ZHANG Y T,TENG Q Z,HE X H,et al. Attention-based 3D CNN with multi-layer features for Alzheimer’s disease diagnosis using brain images[C]//2023 45th Annual International Conference of the IEEE Engineering in Medicine&Biology Society(EMBC).July 24-27,2023.Sydney,Australia.IEEE,2023:1-4.
[30]CHEN L,QIAO H Z,ZHU F.Alzheimer’s disease diagnosis with brain structural MRI using multiview-slice attention and 3D convolution neural network[J].Frontiers in Aging Neuroscience,2022,14:871706.
[31]ZHOU J,WEI Y M,LI X,et al.A deep learning model for early diagnosis of Alzheimer’s disease combined with 3D CNN and video Swin transformer[J].Scientific Reports,2025,15:23311.
基本信息:
中图分类号:TP183;R749.16;R741
引用信息:
[1]刘艺帆,李高平,苗加庆,等.DeMoNet：基于sMRI解耦的多组织融合注意力增强的3D-CNN深度辅助AD诊断神经网络[J].西南民族大学学报(自然科学版)().
基金信息:
国家自然科学基金(项目编号:12271083); 四川省科技项目(项目编号:2024YFFK0362,2024ZYD0087,2026YFHZ0219); 四川省中医药项目(项目编号:2024ZD014); 西南民族大学2025年中央高校基本科研业务费专项资金项目(项目编号:ZYN2025112)的支持
2026-06-16
2026-06-16
2026-06-16