Computerassistierte Segmentierung und Formanalyse von 2975 mammären Mikroverkalkungen mit Hilfe der 7fachen Vergrößerungspräparateradiographie

Abstract

To classify mammary microcalcifications for the prediction of malignancy by using computer-aided analysis of morphometric characteristics. Seven-fold microfocus magnification radiography (direct magnification) was performed on 2975 paraffin embedded microcalcifications. After digitization of the radiographic films and segmentation of the microcalcifications, the morphometric characteristics, such as circumference, surface and polar transformation with determination of the greatest and smallest radius, were measured. Using the classification and regression tree (CART) statistical analysis program, the calcifications were classified by computer on the basis of the morphometric characteristics and the known histological result. Benign conditions were essentially represented by small round calcifications with a greatest radius of < or = 114.2 microm, a smallest radius of > 18.5 microm and a circumference of < or = 462.3 microm. Using > 70 % microcalcifications classified in a patient by the computer as "malignant," CART increases the frequency of malignant conditions, with the diagnostic separation between benign and malignant at an Az value (surface below the ROC curve) of 0.7863. A computer-assisted classification of individual microcalcifications on the basis of morphometric characteristics can supplement the information for a computer assisted mammographic diagnosis (CAD).