Evaluation of the Alzheimer-Type Dementia by Magnetic Resonance Imaging Using Fuzzy Neural Networks

Abstract

A system for evaluating dementia of the Alzheimer type (DAT) based on magnetic resonance (MR) imaging by means of fuzzy neural networks (FNNs) was investigated. The T1-weighted head MR transverse section images were obtained by a routinely performed examination. Nine slices including the thalamus were analyzed for each subject. Each MR image (MRI) was divided into four parts. The ratio of the brain area to the intracranial area was defined as the atrophy ratio. DAT severity was assessed by the Mini-Mental State (MMS) examination administered to each patient, and the results were used as teaching values for the FNN models. To construct the FNN model with high accuracy, MRI-based input variables were examined. Using atrophy ratios of 9 MRIs based on thalamically fiducial images and the corresponding areal or volumetric data as input variables, highly accurate FNN models were constructed that gave an average error of 1.29 points out of 30 in the MMS scores.