Investigation of Methodologies for the Segmentation of Squamous Epithelium from Cervical Histological Virtual Slides

This paper investigates image segmentation methods for the automated identification of Squamous epithelium from cervical virtual slides. Such images can be up to 120Ktimes80K pixels in size. Through investigation a multiresolution segmentation strategy was developed to give the best segmentation results in addition to saving processing time and memory. Squamous epithelium is initially segmented at a low resolution of 2X magnification. The boundaries of segmented Squamous epithelium are further fine tuned at the highest resolution of 40X magnification. Robust texture feature vectors were developed in conjunction with a support vector machine (SVM) to do classification. Finally medical histology rules are applied to remove misclassifications. Results show that with selected texture features, SVM achieved more than 92.1% accuracy in testing. In tests with 20 virtual slides, results are promising.