Optoelectronic morphological processor for cervical cancer screening

Abstract

Pap-smears are slides of cellular material used to screen for cervical cancer. Currently pap-smears are examined manually, a repetitive and tedious task which leads to about 18 % false negative rate (% of abnormal slides going undetected) [1]. Automated pap-smear examination is desirable as both a quality control mechanism for detecting abnormal slides missed by human inspection and as a primary diagnostic cytology screen. Automated screening is challenging since it is a typical example of the “needle-in-a-haystack” problem, where the features of interest are hidden in a vast search area. In a pap-smear one in 10,000 cells screened may be abnormal. Detecting this cell requires high computation power and throughput. Each slide is 2.5 cm x 5.0 cm on a side. Features of interest (the cell nuclei) are on average 10 microns in diameter and sampling at 0.8 μm/pixel, (equivalent to screening the slide with a 20x objective) therefore requires at least 37,000 images of 256x256 pixels to be processed for each slide screened. In this paper we present an optoelectronic implementation of the morphological hit-or-miss transform which can scan each pap-smear slide and detect the regions of interest (ROI) in that slide in under five minutes.