Cytology automation: An overview

© The Papanicolaou smear has been the most effective cancer-screening test ever. It has resulted in a 70% decrease in cervical cancer death over the last 50 years. However, its limitations are well recognized with an average false negative rate for a single smear ranging from 20% to 50% and an irreducible false negative rate of 5% in the best of laboratories.1 About one half of these falsely negative smears lack abnormal cells and represent suboptimal sampling of the cervix. Most of the remaining cases usually have a few abnormal cells that have been missed by the human screener. This major limitation of traditional screening is due to the fallibility of the eye-brain axis of the human observer who is subject to fatigue when viewing literally millions of cells per day (50 000/cells/slide × 60 slides/day = 3 000 000 cells/day). Psychologic habituation, the tendency to not perceive a rare abnormal cell after a long sequence of normals, also plays a role.3 It has long been hoped that the application of automated image analysis to cervical smears would result in a tireless screening device whose accuracy and vigilance would markedly reduce, if not, eliminate false negatives due to screening errors. Such a device could also increase productivity and provide a solution to shortages of trained personnel.