Automated prescreening of conventionally prepared cervical smears: a feasibility study

The feasibility of an automated screening device that will analyze conventionally prepared cervical smears was examined. Consecutive normal and abnormal cervical smears were selected retrospectively and analyzed on a customized microscope imaging workstation system. Specialized image processing, feature extraction, and object classification algorithms were integrated on the imaging workstation to provide an Analysis Score for each slide analyzed. A threshold was set for the Analysis Score, and all slides with a score less than the threshold were classified as normal. Examination of the distribution of the Analysis Score for normal and abnormal cervical smears provided an estimate that 50% of all slides will be sorted as normal, and consequently need no cytologist review. With the Analysis Score threshold set for a 50% sort rate, zero false-negative slides were found in this slide set. This study has shown that a large percentage of slides can be correctly classified as normal while maintaining a high sensitivity to abnormal slides. Given the current workload problems in cytology laboratories, reducing the screening workload by one half will be beneficial. This study has shown that it is now feasible with current technology to provide a clinically useful device to automate screening of conventional cervical smears.