Segmentation of nonconvex regions within uterine cervix images

The National Cancer Institute has collected a large database of uterine cervix images, termed “cervigrams ” for cervical can-cer screening research. Tissues of interest within the cervi-gram, in particular the lesions, are of varying sizes and com-plex, non-convex shapes. The current work proposes a new methodology that enables the segmentation of non-convex re-gions, thus providing a major step forward towards cervigram tissue detection and lesion delineation. The framework transi-tions from pixels to a set of small coherent regions (superpix-els), which are grouped bottom-up into larger, non-convex, perceptually similar regions, utilizing a new graph-cut cri-terion and agglomerative clustering. Superpixels similarity is computed via a combined region and boundary informa-tion measure. Results for a set of 120 cervigrams, manually marked by a medical expert, are shown. Index Terms — cervicography images; segmentation; graph algorithms 1