Automatic detection of Alzheimer disease in brain magnetic resonance images using fractal features

This paper describes a fractal-based processing methodology to detect Alzheimer's disease (AD) in brain magnetic resonance images (MRI). The proposed diagnosis system does not require image preprocessing and segmentation, leading to a simple implementation. The brain MRI is transformed first into a one-dimensional (1-D) signal for faster processing. Then, a three-component feature vector is extracted to characterize the 1-D signal's local and global fractal features. The features include Hurst's exponent and two results from detrended fluctuation analysis (DFA): the scaling exponent and the total fluctuation energy. The validation with ten normal brain MRIs and thirteen abnormal MRIs corresponding to AD led to 100% classification accuracy using support vector machines with a quadratic kernel. It is concluded that the proposed methodology can be as accurate as the best alternative approach while being simpler to implement.