Master thesis presentation
Welcome to the following master thesis presentation:
Chromatin pattern analysis of cell nuclei for improved cervical cancer screening
Presented by Ramin Moshavegh and Babak Ehteshami
Examiner: Andrew Mehnert
The Papanicolaou (Pap) test is the primary screening test for cervical cancer. It involves the microscopic examination of cells sampled from the cervix. Two major factors affect the accuracy of the Pap test. The first is sampling error wherein no diagnostic cells make it on to the slide. The other is interpretation error for reasons including fatigue, inexperience, and habituation. Computer-assisted interpretation can potentially address the issue of interpretation error. The malignancy-associated change (MAC) phenomenon may potentially address sampling error. MACs are subtle sub-visual changes in the appearance of otherwise normal-looking cells from an abnormal Pap slide.
An essential first step in the development of an automated screener, based on MACs, is robust automatic segmentation of free-lying cell nuclei in digitized Pap smear images. This thesis presents and evaluates a fully automated algorithm for robustly detecting and segmenting free-lying cell nuclei in bright-field microscope images of Pap smears. The proposed novel segmentation algorithm makes use of grey-scale annular closings to identify free-lying nuclei-like objects together with marker-controlled watershed segmentation to accurately delineate the nuclear boundaries. The method was evaluated empirically using images digitized from Pap smears sourced from the Regional Cancer Centre in Thiruvananthapuram in India. The results show that the sensitivity and specificity of nucleus detection is 94.71% and 85.30% respectively, and that the accuracy of segmentation, measured using the Dice coefficient, of the detected nuclei is 97.30±1.3%.
This thesis also presents and evaluates a set of novel structural texture features for quantifying and classifying nuclear chromatin patterns in cells on a conventional Pap smear. The experimental results demonstrate the efficacy of the proposed structural approach and that a combination of the structural texture features and conventional features can be used to discriminate between normal and abnormal slides with high accuracy. They also demonstrate that it is possible to detect MACs in Papanicoloau stain (which is not stoichiometric). This in turn suggests the possibility of developing a fully automated Pap smear screener based on MACs.