Image analysis and segmentation of anatomical features of cervix uteri in color space

We propose and verify a method for color-based cluster segmentation of the various tissues from ectocervix. That method uses a simplified compartment-like analysis, aiming for a Gaussian Mixture Model (GMM)-tailored segmentation. The tissues of interest are the cervical canal (CC), the columnar epithelium (CE), the squamous epithelium (SE) and the transformation zone (TZ) the latter known as area where pre-cancer is often found [1]. We used an optimization algorithm (maximum-a priori algorithm or MAP) for bimodal segmentation in normalized RGB color-space, as initially we estimated the deterministic values of CC, TZ, CE and SE as pixel sets in a compartmental-like mode. We assessed the MAP algorithm via automatic segmentation of squamous intraepithelial lesions (SIL) and CC. Our segmentation method is based on the estimates of the GMM boundaries for CC, TZ, and SE and their adjacent area-ratios for healthy ectocervices. We demonstrated a segmentation algorithm for CC and pre-cancer lesion detection that performed with high accuracy.