Today, high-resolution MRI scans are able to reveal even the fine details of brain structure. Several methods have been developed to quantify shape differences specific to scans of diseased brains. We have developed a novel method for quantifying shape information based on multidimensional scaling (MDS), a well-known statistical tool. Multidimensional scaling uses distance measures computed from pair-wise image registration of the training set. Image registration establishes spatial correspondence between scans in order to compare them in the same spatial framework. Our novel method has several advantages, including robustness to errors in registrations. Applying our method to 44 brain MRIs showed clear separation between normal and Alzheimer scans. Using our method as basis for classification between normal and Alzheimer scans yielded better performance results compared with using the volume of hippocampus as basis for classification. We also devised a simple measure derived from the MDS approach that was shown to correlate with the Mini Mental State Examination (MMSE), a well-known cognitive test for Alzheimer's disease.
All Science Journal Classification (ASJC) codes