Mass Contour Extraction in Mammographic Images for Breast Cancer Identification

Abstrac t – Mammography is the most e ffect ive too l no w avai lab le for an ear ly diagnosis o f breast cancer . Ho wever , the de tec tion of cancer signs in mammograms is a d i ff icul t task o wing to the grea t number o f non patho logical struc tures which are also present in the image. I t has been shown t hat in current breas t cancer screenings 10%–25% of the tumors are missed by the radio logis ts . For this reason, a lot o f research is cur rently being done to develop sys tems for Computer Aided Detect ion (CADe). Probably, some causes of the fa lse –negative screening examinat ions are that tumoral masses have varying dimension and ir regular shape, their borders a re o f ten i l l–defined and their contrast i s very lo w, thus making di fficult the discr imination from parenchymal structures . Therefore, in a CADe system a prel iminary segmentat ion procedure has to be implemented in order to separate the mass from background t i ssue. In this way, var ious charac ter i s t ics o f the segmented mass can be evaluated, which may be used in a c lass i f ica t ion s tep to discr iminate patho logi cal and negative cases . In this paper we descr ibe an effec tive algor i thm for massive les ions segmenta tion based on region –gro wing technique and we provide ful l de ta i l s o f the per formance eva lua tion procedure used in this spec i fic context .