Development of Alzheimer's Disease Recognition using Semiautomatic Analysis of Statistical Parameters based on Frequency Characteristics of Medical Images

The paper presents an effective algorithm to analyze MR-images in order to recognize Alzheimer's Disease (AD) which appeared in patient's brain. The features of interest are categorized in Features of the Spatial Domain (FSD's) and Features of the Frequency Domain (FFD's) which are based on the first four statistic moments of the wavelet transform. Extracted features have been classified by a multi-layer perceptron Artificial Neural Network (ANN). Before ANN, the number of features is reduced from 44 to 12 to optimize and eliminate any correlation between them. The contribution of this paper is to demonstrate that by using the wavelet transform number of features needed for AD diagnosis has been reduced in comparison with the previous work. We achieved 79% and 100% accuracy among test set and training set respectively, including 93 MR-images.