Quantitative assessment of pap smear cells by PC-based cytopathologic image analysis system and support vector machine

Abstract

Cytologic screening has been widely used for controlling the prevalence of cervical cancer. Errors from sampling, screening and interpretation, still concealed some unpleasant results. This study aims at designing a cellular image analysis system based on feasible and available software and hardware for a routine cytologic laboratory. Totally 1814 cellular images from the liquid-based cervical smears with Papanicolaou stain in 100x. 200x, and 400x magnification were captured by a digital camera. Cell images were reviewed by pathologic experts with peer agreement and only 503 images were selected for further study. The images were divided into 4 diagnostic categories. A PC-based cellular image analysis system (PCCIA) was developed for computing morphometric parameters. Then support vector machine (SVM) was used to classify signature patterns. The results show that the selected 13 morphometric parameters can be used to correctly differentiate the dysplastic cells from the normal cells (p<0.001). Additionally, SVM classifier has been demonstrated to be able to achieve a high accuracy for cellular classification. In conclusion, the proposed system provides a feasible and effective tool for the evaluation of gynecologic cytologic specimens. © Springer-Verlag Berlin Heidelberg 2007.