Deep and joint learning of longitudinal data for Alzheimer's disease prediction. (June 2020)

• Record Type: Journal Article
• Authors: Lei, Baiying; Yang, Mengya; Yang, Peng; Zhou, Feng; Hou, Wen; Zou, Wenbin; Li, Xia; Wang, Tianfu; Xiao, Xiaohua; Wang, Shuqiang
• Is Part Of: Pattern recognition. Volume 102(2020:Jun.)
• Publication Date: 2020-06
• Subjects: Alzheimer's disease -- Longitudinal scores prediction -- Joint learning -- Correntropy -- Deep polynomial network; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2020.107247
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We propose to build a model using multiple time points to predict longitudinal scores. This model incorporates the feature selection, encoding, and ensemble learning. The selected features are encoded by the ensemble DPN to predict scores using SVR. The missing scores are filled by combing all the previous data. The experiments validate our method's effectiveness in predicting longitudinal scores. Abstract: Alzheimer's disease (AD) is an irreversible and progressive neurodegenerative disease. The close AD monitoring of this disease is essential for the patient treatment plan adjustment. For AD monitoring, clinical score prediction via neuroimaging data is highly desirable since it is able to reveal the disease status, adequately. For this task, most previous studies are focused on a single time point without considering relationship between neuroimaging data (e.g., Magnetic Resonance Imaging (MRI)) and clinical scores at multiple time points. Differing from these studies, we propose to build a framework based on longitudinal multiple time points data to predict clinical scores. Specifically, the proposed framework consists of three parts, feature selection based on correntropy regularized joint learning, feature encoding based on deep polynomial network, and ensemble learning for regression via the support vector regression method. Two scenarios are designed for scores prediction. Namely, scenario 1 uses the baseline data to achieve the longitudinal scores prediction, while scenario 2 utilizes all the previous time points data to obtain the predicted scores at the next time point, which can improve the score prediction's accuracy. Meanwhile, the missing clinical scores at longitudinal multiple time points are imputated to solve the incompleteness of the data. Extensive experiments on the public database of Alzheimer's Disease Neuroimaging Initiative (ADNI) demonstrate that our proposed framework can effectively reveal the relationship between clinical score and MRI data and outperforms the state-of-the-art methods in scores prediction.