Application of malignancy-associated changes of the cervical epithelium in a hierarchic classification concept.

For almost 10 years malignancy-associated changes (MAC) have been consistently found by means of high resolution image analysis of apparently normal uterine cervical cells. This study was performed on Feulgen-stained monolayer specimens from 55 healthy women and 19 patients suffering from various stages of cervical neoplasia. About 200 epithelial cells and 30 trout red blood cells in each case were measured by image cytometry at a spatial resolution of 0.27 micron. The philosophy of classification is based on the hierarchic stepwise definition of 'truly normal' specimens by removal of all suspicious specimens, characterized by the presence of atypical cells or subvisually altered cell populations. The suitability of MAC-sensitive classifiers, in combination with classifiers for the recognition of atypical cells, was demonstrated. The demands of automated prescreening were met by several decision tree procedures (95-100% sensitivity, 85-95% specificity). By focussing on the MAC evidence a new risk group from seemingly healthy women may be defined.