Real time mass classification for mammographic images: a Driven CADx scheme
Computer-Aided Diagnosis schemes have been proposed aiming at working as a second image analysis in mammography. Experienced radiologists however tend to be more assertive to such schemes in assisting their interpretation rather than their ability to detect suspicious signals. This work characterizes a simplified version of a mammography CADx scheme we had previously developed directed to digitized film, now however, aimed specifically at classifying breast nod-ules marked as regions of interest on digital images. This “driven” CADx scheme indicates promptly whether the selected nodule is normal or suspicious. Results from tests with different mammograms sets – one with large number of images selected from DDSM database, for training, testing and validation of classification parameters, and other with direct digital images from In-Breast database – registered similar rates regarding sensitivity, specificity and accuracy (83%, 67% and 72%, respectively), using attributes associated to contour, density and texture. A third test performed with radiologists analyzing digital mammograms from particular FFDM unit indicated that the Driven CADx scheme had positive influence on their final diagnoses, always increasing accuracy rates. This promising result allows setting this software as useful tool to the radiologist analysis of masses in digital mammography, running with any operational system, or even online.