Ellipse test for the reduction of false positive signals in automated cytology.

One of the major problems in automated cytology is the elimination of false positive 'abnormal cell' alarms caused by objects such as overlapping cell pairs, leukocyte clusters, etc. The paper describes an algorithm for the separation of images of abnormal cell nuclei and non=nuclear objects in computer image analysis systems for automated cytology. The algorithm involves the measurement of the agreement between the object outline and a computer ellipse of equal area, aspect ratio and orientation. Results obtained with the CERVISCAN experimental computer image analysis system show that the algorithm gives good discrimination between abnormal cell nuclei and typical non-nuclear objects found in cervical scrape specimens prepared specially for automated analysis.