基于聚类过采样和结构稀疏表达的ADHD多模态MRI融合分类

项目名称： 基于聚类过采样和结构稀疏表达的ADHD多模态MRI融合分类

项目编号： No.61300073

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 黄惠芳

作者单位： 北京交通大学

项目金额： 23万元

中文摘要： 注意缺陷多动障碍(ADHD)的准确诊断和及时治疗日益受到社会和医学界的广泛关注。利用磁共振影像(MRI)技术研究神经精神疾病所取得的研究成果极大地促进了神经精神疾病的客观诊断。目前基于磁共振神经影像的ADHD分类是当前的研究热点之一，但是分类性能很不理想，有很大的提升空间。因此，本项目研究基于聚类过采样和结构稀疏表达的ADHD多模态MRI特征融合分类。主要研究结构MRI的局部脑区纹理特征提取和静息状态fMRI频率特定的功能连接特征提取；针对样本分布不平衡问题，通过基于聚类的过采样方法构建类内子块，研究考虑训练词典结构的结构稀疏表达进行ADHD分类，并用于选出有利于分类的结构特征和功能特征；最后利用随机矩阵以少量投影保留原始信息的特点来实现特征级融合，提升分类性能。本项目有助于推动MRI神经影像技术对ADHD疾病的客观诊断，具有重要的基础研究意义和临床应用价值。

中文关键词： 磁共振影像；特征提取；模式分类；不平衡数据；神经精神疾病

英文摘要： The diagnosis and treatment of attention deficit hyperactivity disorder (ADHD) has increasingly gained more attention from society and medical field. The research findings of neuropsychiatric disorders based on magnetic resonance imaging technology have greatly advanced the objective diagnosis of neuropsychiatric disorders. ADHD classification based on magnetic resonance neuroimaging is one of the research hot topics. However, the performance of classification is not satisfactory and has big room for improvement. Therefore, this project aims at multimodal MRI feature fusion classification of ADHD using cluster-based oversampling and structured sparse representation. The main works include three aspects: (1) extracting the texture features of localized brain regions from structural MRI data and frequency-specific functional connectivity features from rest state fMRI data; (2) against the problem of imbalanced samples, using cluster-based oversampling to construct the blocks in the same class and proposing structured sparse representation with training dictionary structure for ADHD classification. This classification model is used to select the discriminative structural and functional features; (3) implementing the feature-level fusion of structural and functional features to boost the classification performance e

英文关键词： Magnetic Resonance Imaging；Feature Extraction；Pattern Classification；Imbalanced Data；Neuropsychiatry