false positive reduction
A false positive is a region being normal tissue but interpreted by the automatic algorithm as a suspicious one. The so called false positive reduction algorithms try to solve this drawback, i.e. given a Region of Interest (RoI) – a sub-image containing the suspicious region – the aim is to validate whether it contains a real lesion or it is only a region depicting normal parenchyma.
We proposed several approaches to perform mass false positive reduction by using the textural properties of the masses. We studied the use of Principal Component Analysis and Local Binary Patterns to characterise micro-patterns (i.e. edges, lines, spots, flat areas) and preserve at the same time the spatial structure of the masses. Our works were the first attempting to use LBP in the field of mammographic mass detection. We showed that using a LBP characterization and Support Vector Machines to classify the ROIs between real masses and normal parenchyma we were able to improve the results on mass false positive reduction.