Trials with the cerviscan experimental prescreening device on polylysine-prepared slides.

A blind trial was carried out to evaluate the performance of the Cerviscan experimental automated cytology prescreening system on cervical scrape specimens prepared by a suspension-polylysine-hematoxylin preparation method. One hundred twenty-nine specimens of various cytologic gradings were automatically scanned, and objects detected as "suspect" were examined and classified. The overall results of the trial showed a false-negative specimen rate of 4% and a false-positive specimen rate of 18%. The mean number of false-positive signals from normal specimens was 24 per 20,000-cell sample. Major problems were the number of false-positive signals caused by the hematoxylin stain and the number of specimens rejected as inadequate by the scanner. A random partition test was carried out to estimate the ROC performance curve for the system.