Exploring breast cancer diagnosis with fractal analysis and classification methods

Screening and diagnostic mammography are the most effective tools available for detection and diagnosis of breast cancer. In the last decade many techniques based upon measures of the shape of the contours of breast masses are been developed to investigate the nature of lesions between malignant tumours and benign masses. This paper presents methods for statistical analysis on a data set of 192 contours of breast masses. Results of these analysis lead to levels of accurate prediction in 90% of the cases, overcoming 98% for the diagnosis of malignant lesions. In this study we applied multivariate statistical techniques for examining relationships among more variables at the same time. We used in addition to the shape factors of contour masses also the age of the patients at the time of mammography, using both ROC analysis and segmentation analysis through Classification and Regression Tree.