Fractal-based arteriovenous malformations detection in brain magnetic resonance images

A new fractal-based methodology to detect cerebral arteriovenous malformations (AVM) in brain magnetic resonance images (MRI) is presented. First, the MRI is preprocessed to emphasize edges. Then, the result is split into right and left brain hemisphere components that are converted to one-dimensional signals, for which the Hurst's exponent, the scaling exponent of detrended fluctuation analysis (DFA) and the energy of DFA fluctuations are computed to form a six-component feature vector. Finally, the vector is classified by a support vector machine (SVM). Using ten-fold cross validation and a set of twenty eight normal and twenty eight MR images of patients affected by AVMs, the classification of the corresponding feature vectors by the SVM achieved an accuracy of 98.26%, with a sensitivity of 98.82% and a specificity of 97.69%.