Fractal analysis for the viewpoint invariant classification of celiac disease

In this paper we present a viewpoint invariant method using fractal features for the automatic classification of celiac disease. First the endoscopic images of the duodenum are pre-filtered using the MR8 filter bank. Subsequently we compute the local fractal dimension of each pixel of the filter responses. After finding representative combinations of the local fractal dimensions for images with or without celiac disease, we generate frequency histograms that count how often each representative combinations occur as nearest neighbor for each image (as nearest neighbor to the local fractal dimensions of the filter responses for each pixel position of the images). We use this histograms for the classification of the images. We will see that the result of the proposed method outperforms the approach, our method is based on, as well as other methods for the classification of medical imagery.