Muti-class diagnosis model and dynamic screening framework for dementia based
on longitudinal neuroimaging data 
   Lead Investigator:    Hongmei Yu
   Institution      :    Shanxi Medical University, China
   E-Mail           :    yu@sxmu.edu.cn
   Proposal ID      :    1507



   Proposal Description:
   The development of neuroimaging provides a new technology for the early diagnosis of mild cognitive impairment (MCI) and Alzheimer's disease (AD), which is of great significance for the early prevention and treatment of AD. In real clinical applications, given neuroimaging data of an subject, he or she      can be categorized into one of the following status, such as normal control (NC), stable MCI (sMCI), progressive MCI (pMCI) and AD. This obviously belongs to a multi-class classification problem in statistics. The AD multi-classification diagnosis models using machine learning have low performance and longitudinal neuroimaging features selection and fusion need to be improved. Based on binary diagnosis model between NC and AD using the single classifier and individual dynamic risk prediction of AD using longitudinal cognitive measures, this study aims: (1) to explore feature extraction, feature selection and feature fusion for longitudinal MRI and PET neuroimaging data by convolutional neural networks and multiple kernel learning respectively using the public database ADNI (2) to build the personalized AD multi-classification diagnosis model by data level, classifier level, and combination level ensemble learning techniques and to compare classification performance of different classifiers by external data validation (3) to establish a dynamic screening framework for cognitive decline among the community-dwelling elderly by integrating neuroimaging data with widely available, cost-effective and non-invasive cognitive measures.