Exploring Ranklets Performances in Mammographic Mass Classification using Recursive Feature Elimination

The ranklet transform is a recently developed image processing technique characterized by a multi-resolution and orientation-selective approach similar to that of the wavelet transform. Yet, differently from the latter, it deals with pixels´ ranks rather than with their gray-level intensity values. In this work, the ranklet coefficients resulting from the application of the ranklet transform to regions of interest (ROIs) found on breast radiographic images are used as classification features to determine whether ROIs contain mass or normal tissue. Performances are explored recursively eliminating some of the less discriminant ranklet coefficients according to the cost function of a support vector machine (SVM) classifier. Experiments show good classification performances (A_z values of 0.976 plusmn 0.003) even after a significant reduction of the number of ranklet coefficients.