Manifold Embedding Induced by Multidimensional Scaling and Its Application to Alzheimer's Disease and Mild Cognitive Impairment

Quantifying shape information related to a disease such as Alzheimer's disease (AD) is an important area of brain research. Neuroimaging data are highly dimensional and are thus cumbersome to analyze. Manifold learning techniques, which find a low-dimensional representation for highdimensional data, have been applied to brain magnetic resonance imaging (MRI). A shape quantification method based on multidimensional scaling (MDS), a well-known manifold learning technique, was proposed. The method successfully distinguished between AD and normal patients. We extended the MDS-based quantification method by 1) applying it to distinguish patients with mild cognitive impairment (MCI) from normal patients, and 2) showing the effectiveness of the induced low-dimensional embedding space in predicting key clinical variables such as mini-mental-state exam scores and clinical diagnosis by using a standard multiple linear regression. Distinguishing patients with MCI from normal patients is also important as it is related to early detection of AD. We were able to 1) classify not only AD/normal patients but also MCI/normal patients better than we could with the traditional classification based on the hippocampus volume, and to 2) show good statistical power for predicting key clinical variables. Our test group consisted of 25 normal, 25 AD, and 25 MCI patients.