Ordinal Ranking for Detecting Mild Cognitive Impairment and Alzheimer’s Disease Based on Multimodal Neuroimages and CSF Biomarkers

Abstract

Early diagnosis of Alzheimer's disease (AD) based on neuroimaging and fluid biomarker data has attracted a lot of interest in medical image analysis. Most existing studies have been focusing on two-class classification problems, e.g., distinguishing AD patients from cognitive normal (CN) elderly or distinguishing mild cognitive impairment (MCI) individuals from CN elderly. However, to achieve the goal of early diagnosis of AD, we need to identify individuals with AD and MCI, especially MCI individuals who will convert to AD, in a single setting, which essentially is a multi-class classification problem. In this paper, we propose an ordinal ranking based classification method for distinguishing CN, MCI non-converter (MCI-NC), MCI converter (MCI-C), and AD at an individual level, taking into account the inherent ordinal severity of brain damage caused by normal aging, MCI, and AD, rather than formulating the classification as a multi-class classification problem. Experiment results indicate that the proposed method can achieve a better performance than traditional multi-class classification techniques based on multimodal neuroimaging and CSF biomarker data of the ADNI.