Fractal dimension and high order statistics of spectral energy distribution as features for pathology detection in brain MR images

A new methodology to detect pathologies in human brain magnetic resonance (MR) images is investigated. It is based on edge extraction in the Hilbert domain and subsequent analysis by means of fractal dimension and spectral energy distribution high order statistics. The technique is particularly suitable for pathologies characterized by bright structures in the MR images as do Glioma and Metastatic bronchogenic carcinoma. When classifying normal versus abnormal images dues to these two pathologies, ANOVA statistics show that the suggested features have strong between-classes differences, and the obtained classification accuracy by support vector machines is 99.9%±0.006. In comparison, applying a standard feature extraction technique based on the discrete wavelet transform (DWT) and principal component analysis (PCA) yielded 85.2%±0.05 accuracy.