Histology-based oral lesion classification

Paper in proceeding, 2012

A computer aided diagnosis (CADx) system for classification of oral cavity lesions from histological images based upon image analysis and pattern recognition has been developed. The aim was to discriminate normal tissue against two of the common and potentially precancerous lesions, Oral Lichen Planus and Oral Submucous Fibrosis, using SVM and kNN classifiers. We proposed to investigate the histogram-based properties of the tissue as discriminating features. Also, two color representation modalities (RGB and HSV) were used to evaluate their discriminative power for analysis of histological images. Relying only on the histogram features, the overall classification accuracy was 83.7% with sensitivity and specificity of 89% and 74%, respectively. Employing the color systems, the best result was achieved in the HSV system (78% accuracy)