Yan Jiexi, Deng Cheng, Luo Lei, Wang Xiaoqian, Yao Xiaohui, Shen Li, Huang Heng
School of Electronic Engineering, Xidian University, Xi'an, China.
Electrical and Computer Engineering, University of Pittsburgh, Pittsburgh, PA, United States.
Front Neurosci. 2019 Jul 10;13:668. doi: 10.3389/fnins.2019.00668. eCollection 2019.
Alzheimer's disease (AD) is a severe type of neurodegeneration which worsens human memory, thinking and cognition along a temporal continuum. How to identify the informative phenotypic neuroimaging markers and accurately predict cognitive assessment are crucial for early detection and diagnosis Alzheimer's disease. Regression models are widely used to predict the relationship between imaging biomarkers and cognitive assessment, and identify discriminative neuroimaging markers. Most existing methods use different matrix norms as the similarity measures of the empirical loss or regularization to improve the prediction performance, but ignore the inherent geometry of the cognitive data. To tackle this issue, in this paper we propose a novel robust matrix regression model with imposing Wasserstein distances on both loss function and regularization. It successfully integrate Wasserstein distance into the regression model, which can excavate the latent geometry of cognitive data. We introduce an efficient algorithm to solve the proposed new model with convergence analysis. Empirical results on cognitive data of the ADNI cohort demonstrate the great effectiveness of the proposed method for clinical cognitive predication.
阿尔茨海默病（AD）是一种严重的神经退行性疾病，它会在时间轴上逐渐恶化人类的记忆、思维和认知能力。如何识别信息丰富的表型神经影像标志物并准确预测认知评估结果，对于阿尔茨海默病的早期检测和诊断至关重要。回归模型被广泛用于预测影像生物标志物与认知评估之间的关系，并识别具有判别力的神经影像标志物。大多数现有方法使用不同的矩阵范数作为经验损失或正则化的相似性度量来提高预测性能，但忽略了认知数据的内在几何结构。为了解决这个问题，在本文中我们提出了一种新颖的鲁棒矩阵回归模型，该模型在损失函数和正则化中都引入了 Wasserstein 距离。它成功地将 Wasserstein 距离整合到回归模型中，能够挖掘认知数据的潜在几何结构。我们引入一种高效算法来求解所提出的新模型，并进行了收敛性分析。阿尔茨海默病神经影像倡议（ADNI）队列的认知数据实证结果表明，所提出的方法在临床认知预测方面具有显著效果。
