Hybrid segmentation, characterization and classification of basal cell nuclei from histopathological images of normal oral mucosa and oral submucous fibrosis

• Published in: Expert systems with applications 2012, Vol.39 (1), p.1062-1077
• Main Authors: Muthu Rama Krishnan, M., Chakraborty, Chandan, Paul, Ranjan Rashmi, Ray, Ajoy K.
• Format: Article
• Language: English
• Subjects: Color deconvolution; Fuzzy divergence; Gradient vector flow; Oral submucous fibrosis; Parabola fitting; Pattern classification; Unsupervised feature selection; Zernike moments
• ISSN: 0957-4174; 1873-6793
• DOI: 10.1016/j.eswa.2011.07.107

► In this study we provide a quantitative analysis of basal cell nuclei to discriminate normal and OSF groups. ► Morphology and texture of basal nuclei are assumed to vary during malignant transformation. ► Hybrid segmentation algorithm is used to segment the basal cell nuclei. ► An extracted morphology and texture feature provides higher accuracy (99.66%) using SVM. This work presents a quantitative microscopic approach for discriminating oral submucous fibrosis (OSF) from normal oral mucosa (NOM) in respect to morphological and textural properties of the basal cell nuclei. Practically, basal cells constitute the proliferative compartment (called basal layer) of the epithelium. In the context of histopathological evaluation, the morphometry and texture of basal nuclei are assumed to vary during malignant transformation according to onco-pathologists. In order to automate the pathological understanding, the basal layer is initially extracted from histopathological images of NOM ( n = 341) and OSF ( n = 429) samples using fuzzy divergence, morphological operations and parabola fitting followed by median filter-based noise reduction. Next, the nuclei are segmented from the layer using color deconvolution, marker-controlled watershed transform and gradient vector flow (GVF) active contour method. Eighteen morphological, 4 gray-level co-occurrence matrix (GLCM) based texture features and 1 intensity feature are quantized from five types of basal nuclei characteristics. Afterwards, unsupervised feature selection method is used to evaluate significant features and hence 18 are obtained as most discriminative out of 23. Finally, supervised and unsupervised classifiers are trained and tested with 18 features for the classification between normal and OSF samples. Experimental results are obtained and compared. It is observed that linear kernel based support vector machine (SVM) leads to 99.66% accuracy in comparison with Bayesian classifier (96.56%) and Gaussian mixture model (90.37%).