College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China.
Department of Radiology and BRIC, University of North Carolina at Chapel Hill, NC 27599, USA.
Med Image Anal. 2019 Apr;53:111-122. doi: 10.1016/j.media.2019.01.007. Epub 2019 Jan 30.
Alzheimer's disease (AD) is a neurodegenerative disorder characterized by progressive impairment of memory and other cognitive functions. Currently, many multi-task learning approaches have been proposed to predict the disease progression at the early stage using longitudinal data, with each task corresponding to a particular time point. However, the underlying association among different time points in disease progression is still under-explored in previous studies. To this end, we propose a multi-task exclusive relationship learning model to automatically capture the intrinsic relationship among tasks at different time points for estimating clinical measures based on longitudinal imaging data. The proposed method can select the most discriminative features for different tasks and also model the intrinsic relatedness among different time points, by utilizing an exclusive lasso regularization and a relationship induced regularization. Specifically, the exclusive lasso regularization enables partial group structure feature selection among the longitudinal data, while the relationship induced regularization efficiently introduces the relationship information from data to guide knowledge transfer. We further develop an efficient optimization algorithm to solve the proposed objective function. Extensive experiments on both synthetic and real datasets demonstrate the effectiveness of our proposed method. In comparison with several state-of-the-art methods, our proposed method can achieve promising performance for cognitive status prediction and also can help discover disease-related biomarkers.
阿尔茨海默病（AD）是一种神经退行性疾病，其特征是记忆和其他认知功能的进行性损害。目前，已经提出了许多多任务学习方法，以便使用纵向数据在早期预测疾病的进展，每个任务对应于特定的时间点。然而，在以前的研究中，疾病进展过程中不同时间点之间的潜在关联仍未得到充分探索。为此，我们提出了一种多任务排他性关系学习模型，以自动捕捉不同时间点任务之间的内在关系，从而基于纵向成像数据估计临床指标。所提出的方法可以选择不同任务的最具判别力的特征，并且可以通过利用排他性 LASSO 正则化和关系诱导正则化来模拟不同时间点之间的内在相关性。具体来说，排他性 LASSO 正则化可以在纵向数据中实现部分组结构特征选择，而关系诱导正则化可以有效地从数据中引入关系信息来指导知识转移。我们进一步开发了一种有效的优化算法来解决所提出的目标函数。在合成和真实数据集上的广泛实验证明了我们提出的方法的有效性。与几种最先进的方法相比，我们提出的方法可以在认知状态预测方面取得有希望的性能，并且还可以帮助发现与疾病相关的生物标志物。
