High-dimensional pattern classification was applied to baseline and multiple follow-up MRI scans of the Alzheimer's Disease Neuroimaging Initiative (ADNI) participants with mild cognitive impairment (MCI), in order to investigate the potential of predicting short-term conversion to Alzheimer's Disease (AD) on an individual basis. MCI participants that converted to AD (average follow-up 15 months) displayed significantly lower volumes in a number of grey matter (GM) regions, as well as in the white matter (WM). They also displayed more pronounced periventricular small-vessel pathology, as well as an increased rate of increase of such pathology. Individual person analysis was performed using a pattern classifier previously constructed from AD patients and cognitively normal (CN) individuals to yield an abnormality score that is positive for AD-like brains and negative otherwise. The abnormality scores measured from MCI non-converters (MCI-NC) followed a bimodal distribution, reflecting the heterogeneity of this group, whereas they were positive in almost all MCI converters (MCI-C), indicating extensive patterns of AD-like brain atrophy in almost all MCI-C. Both MCI subgroups had similar MMSE scores at baseline. A more specialized classifier constructed to differentiate converters from non-converters based on their baseline scans provided good classification accuracy reaching 81.5%, evaluated via cross-validation. These pattern classification schemes, which distill spatial patterns of atrophy to a single abnormality score, offer promise as biomarkers of AD and as predictors of subsequent clinical progression, on an individual patient basis.

Spatial patterns of brain atrophy in mild cognitive impairment (MCI) and Alzheimer's disease (AD) were measured via methods of computational neuroanatomy. These patterns were spatially complex and involved many brain regions. In addition to the hippocampus and the medial temporal lobe gray matter, a number of other regions displayed significant atrophy, including orbitofrontal and medial-prefrontal grey matter, cingulate (mainly posterior), insula, uncus, and temporal lobe white matter. Approximately 2/3 of the MCI group presented patterns of atrophy that overlapped with AD, whereas the remaining 1/3 overlapped with cognitively normal individuals, thereby indicating that some, but not all, MCI patients have significant and extensive brain atrophy in this cohort of MCI patients. Importantly, the group with AD-like patterns presented much higher rate of MMSE decline in follow-up visits; conversely, pattern classification provided relatively high classification accuracy (87%) of the individuals that presented relatively higher MMSE decline within a year from baseline. High-dimensional pattern classification, a nonlinear multivariate analysis, provided measures of structural abnormality that can potentially be useful for individual patient classification, as well as for predicting progression and examining multivariate relationships in group analyses.

This paper presents a method for classification of structural brain magnetic resonance (MR) images, by using a combination of deformation-based morphometry and machine learning methods. A morphological representation of the anatomy of interest is first obtained using a high-dimensional mass-preserving template warping method, which results in tissue density maps that constitute local tissue volumetric measurements. Regions that display strong correlations between tissue volume and classification (clinical) variables are extracted using a watershed segmentation algorithm, taking into account the regional smoothness of the correlation map which is estimated by a cross-validation strategy to achieve robustness to outliers. A volume increment algorithm is then applied to these regions to extract regional volumetric features, from which a feature selection technique using support vector machine (SVM)-based criteria is used to select the most discriminative features, according to their effect on the upper bound of the leave-one-out generalization error. Finally, SVM-based classification is applied using the best set of features, and it is tested using a leave-one-out cross-validation strategy. The results on MR brain images of healthy controls and schizophrenia patients demonstrate not only high classification accuracy (91.8% for female subjects and 90.8% for male subjects), but also good stability with respect to the number of features selected and the size of SVM kernel used.

No abstract