基于纵向多模态数据的阿尔茨海默症早期诊断方法研究

Alzheimer's disease (AD), also called senile dementia, is a progressive age-related neurodegenerative disease. The medical research on AD has proved that the correct and early diagnosis of AD is great important for patients to receive an effective early intervention and treatment for AD. But the state-of-the-art methods of early diagnosis of AD, based on either the baseline data or the longitudinal data, fail to reach the satisfied requirement for computer-aided early diagnosis of AD, mainly because of the inadequate numbers of samples, the incomplete modalities, the block-wise missing data and without enough consideration of quantized assessment of cognitive function. Hence, the novel framework of the early diagnosis of AD based on the longitudinal multimodality data is proposed to overcome these difficulties. That is, the samples and their features will be effectively selected via the multi-task transfer learning and the multi-view multi-task transfer learning respectively to compensate the inadequate numbers of samples and the block-wise missing data. Then, an SVM based on a multi-view domain transfer learning will classify the samples into MCI-C type or MCI-NC type and realize the regression of the quantized assessment under adequate consideration of cognitive function. The project aims at providing the effective early diagnosis of AD to the physicians and laying the foundations for computer-aided early diagnosis of AD.

阿尔茨海默症(AD)又称老年性痴呆，是一种进行性非可逆神经系统退行性疾病。医学研究表明AD的早期诊断对它的干预和治疗非常重要。由于样本数量少、模态缺失、数据中断以及未充分考虑神经心理学量表评估，造成目前基于基线数据和纵向数据的AD早期诊断方法效果均不尽人意。为解决该类问题，本项目提出研究新的基于纵向多模态数据的AD早期诊断方法。首先采用多任务迁移学习的样本选择和多视角多任务迁移学习的特征选择方法解决样本不足、模态缺失和数据中断的问题。其次，综合考虑纵向多模态数据的分类和神经心理学量表分值的回归，最终实现可靠的早期诊断，从而为有效的计算机辅助AD早期诊断提供技术支撑。

阿尔茨海默症(AD)又称老年性痴呆，是一种进行性非可逆神经系统退行性疾病。医学研究表明AD的早期诊断对它的干预和治疗非常重要。由于样本数量少、模态缺失、数据中断以及未充分考虑神经心理学量表评估，造成目前基于基线数据和纵向数据的AD早期诊断方法效果均不尽人意。为解决该类问题，本项目提出研究新的基于纵向多模态数据的AD早期诊断方法。本项目研究内容包括：研究了基于偏移场校正的超体素分割和分类方法，为AD和健康人群的分类奠定了基础；开展了基于功能核磁共振图像特定脑区的分析和分类研究，为大脑精神类疾病的治疗提供了佐证；开展了基于个体化灰质形态学网络的计算机辅助诊断研究，并同步定位了作用于分类的关键脑区，为大脑精神类疾病的研究提供了潜在的影像学标记。本项目实现了可靠的早期诊断，为有效的计算机辅助AD早期诊断提供了技术支撑。