Fractal Feature Analysis and Classiication of Ms Lesions

We present a novel approach to detect and classify multiple sclerosis lesions in magnetic resonance images by using the concept of fractional Brownian motion. Estimating the local fractal dimension by means of its multiscale fractional Brownian Feature Vector (BFV) allows to distinguish between diierent types of brain tissue with regard to the human perception of roughness and statistical self-similarity of the underlying structure. While the fractal dimension alone does not provide suucient discriminatory power to classify textures, the ooset of the BFV in combination with the dimension is able to detect and distinguish lesions from other tissues.