Probabilistic Segmentation Methods for Early Detection of Uterine Cervical Cancer

Abstract

Uterine Cervical Cancer is one of the prevalent forms of cancer in women worldwide. Most cases of cervical cancer can be prevented through screening programs aimed at detecting precancerous lesions. In this paper, novel methods have been proposed for automated probabilistic image segmentation of cervical cancer. The detection of cervical lesions is an important issue in image processing because it has a direct impact on surgical planning. We examined the segmentation accuracy based on a validation metric against the estimated composite latent gold standard, which was derived from several experts’ manual segmentations. The distribution functions of the lesion and control pixel data were parametrically assumed to be a mixture of probability distributions with different shape parameters. We also estimated the corresponding receiver operating characteristic (ROC) curve over all possible decision thresholds. The automated segmentation yielded satisfactory accuracy with protean optimal thresholds.

Keywords: Segmentation, Clustering, Gaussian Mixture Model

View PDF