Local and Global Fractal Behaviour in Mammographic Images

Breast cancer is one of the leading causes of cancer in women. Most studies attempt to perform segmentation of tumors highlighted in mammogaphic images, or analysis of the contours of tumors for classification purposes. Successful segmentation and classification of tumors can assist physicians in revealing suspicious regions or masses, or differentiating malignant from benign tumors in the mammogram. However, relevant studies do not focus on the tumor surface statistics for the purpose of clustering or classification. In this work, we present a statistical, fractal-based approach, for the analysis of annotated tumors, reduced from the DDSM database. Using local and global fractal properties, obtained from the tumor surface, we show that malignant and benign tumors from are separable in an appropriate feature space. K-means-based clustering is performed, showing the efficacy of the method.