Computer classification of stellate carcinomas and fibroadenomas using digitized mammograms

Abstract

T h e final goal of this research is to deslgn a n d implement a computer based system capable of distinguishing between a number of different malignant and benign breast tumors. I t is envisioned that, by incorporat ing computer classification techniques along with patient age and clinical history into a n expert system, the f h a l instrument will be used as a screening device to aid the radiologist In classifying the tumor type noninvasively, potentially without the need fo r biopsy. This paper will describe the first pliase of this approach which includes the incorporat ion of three features to ullow computer differentiation between stellate carcinonras and fibroadenomas. The scanning nicdla was tlie digitized ininge of a film-screen mammogranl . A user friciidly systeni was designed a n d developed which used circularity, su r face roughness, and boundary rougliness as the features to identify those character is t ics in thc digital iniugc which indicate malignancy.