Abstract: In recent years, multimodal Magnetic Resonance Imaging (MRI) has demonstrated significant advantages in brain disease diagnoses and brain network analyses. However, challenges remain in effectively optimizing and correlating multimodal data such as resting-state functional connectivity and white matter fiber structure using resting-state functional MRI (rs-fMRI) and Diffusion Tensor Imaging (DTI), respectively, while extracting low-dimensional brain network features with strong topological representation capabilities. This study proposes a joint optimization framework to address the issues of feature optimization and topological information capture and utilization in multimodal brain network analysis rs-fMRI/DTI. First, to mitigate feature distribution shifts and modal heterogeneity, a SPAMS-based multimodal dictionary learning data enhancement strategy is proposed. By jointly optimizing functional connectivity brain networks and diffusion tensor brain networks, shared sparse dictionaries are constructed to generate anatomically functional, consistent, and enhanced data, thereby improving intergroup similarity and feature quality. Second, to effectively capture complex topological information in brain networks, a Riemannian Manifold-constrained Loss Auto-Encoder (RM-Loss AE) is introduced. This model constructs a feature space as a positive definite matrix manifold and incorporates optimization reconstruction losses such as the log-Euclidean metric. Comprehensive experiments on the Alzheimer's Disease Neuroimaging (ADNI) and Autism Brain Imaging Data Exchange (ABIDE-Ⅱ) datasets demonstrate that the proposed method significantly enhances key metrics including feature separability, classification performance, and the coupling strength between rs-fMRI and DTI modalities. This study establishes a novel paradigm for multimodal brain network representation learning, advancing its application in precision medicine.

Key words: data enhancement, manifold optimization, brain disease classification, Alzheimer′s Disease（AD）, autism

摘要： 近年来，多模态磁共振成像(MRI)技术在脑疾病诊断与脑网络分析中展现出显著优势，但如何在磁共振数据静息态功能连接（rs-fMRI）、弥散张量成像（DTI）等多模态进行数据优化和关联化，并提取具有强拓扑表征能力的低维脑网络特征的研究仍面临挑战。针对多模态脑网络rs-fMRI/DTI分析中特征优化、拓扑信息捕获和利用的问题，提出联合优化框架。首先，为缓解特征分布偏移与模态异质性，提出基于SPAMS的多模态字典学习数据增强策略，通过联合优化功能连接的脑网络、弥散结构脑网络，构建共享稀疏字典，生成解剖-功能一致性增强数据，提升组间相似性与特征质量；其次，为有效捕获脑网络复杂拓扑信息，提出黎曼流形约束损失自编码器（RM-Loss AE），将特征空间建模为正定矩阵流形，引入对数欧氏度量等优化重构损失。在阿尔茨海默病神经影像学（ADNI）和自闭症脑成像数据交换-Ⅱ(ABIDE-Ⅱ)数据集上的实验结果表明，所提方法显著提升了特征可分性、分类性能以及rs-fMRI与DTI模态间的耦合强度等指标，为多模态脑网络表征学习提供了新范式，推动其在精准医疗中的应用。

关键词: 数据增强, 流形优化, 脑疾病分类, 阿尔茨海默病, 自闭症

CLC Number: 

• TP391