Department of Computer Engineering
S E M I N A R
Segmentation and Classification of Cervical Cell Images
Computer Engineering Department
Cervical cancer is a preventable disease and the dysplasia it causes can be scanned by using a pap smear test. It can be beneficial to develop a computer-assisted diagnosis system to make the pap smear test robust and widespread. For example, the cells in a given pap smear test image can be sorted starting from normal to abnormal by the aid of such a system and the cells coming before a selected cell which is found to be normal can be skipped while deciding on the cancer degree of the image. The most fundamental part of the proposed system is the segmentation of nuclei and cytoplasm in cervical cell images. After segmentation step, a number of features for each cell can be extracted and based on these features; cells can be sorted according to their abnormality. In this study, we concentrated on the segmentation of the cell nuclei in pap test images. First, markers on the nuclei are found by using mathematical morphology operations. Based on the obtained markers, marker-based watershed segmentation and balloon snake model are applied to find the nuclei contours in a data set consisting of cervical cell images. The data set is composed of six classes ranging according to the dysplasia degree of the cells. The results are evaluated according to the relative distance error measure, and the strengths and weakness of the methods are discussed.
DATE: 6 April, 2009, Monday@ 15:40
PLACE: EA 502