Automatic Classification of Leukoplakia on Vocal Folds Using Color Texture Features

Abstract

In this paper we present novel approaches for the computer aided diagnosis of leukoplakia on the vocal folds based on video endoscopic images of the larynx. The approaches applied for the classification of the vocal fold surface tissue are texture analysis methods, which have been enhanced to use spatial information as well as color information in one combined color texture approach. The multispectral approaches used are based on sumand difference histograms as well as statistical geometrical features. These novel features are applied on manually segmented and preclassified regions and subregions of the vocal folds. It can be shown that the use of these combined color and texture features is dominant to approaches, which make solely use of either color or texture features. 1 Motivation and Problem Definition Leukoplakia as premalignant lesion of the larynx can be regarded as a patchy buildup of tissue on the true vocal folds which has a leathery, white, inhomogeneous ’crusty’ appearance. These lesions can be isolated or bilateral, and are mostly asymmetrical regarding appearance and effect. With leukoplakia, a keratinizing cell layer develops as a whitish plaque on the surface of the vocal fold. As this layer enlarges, it may distort the glottal wave. As a result the voice becomes hoarse, and persistent coughing and throat clearing may follow. This abnormal mass on the vocal folds must be treated as a precancerous lesion, in the majority of cases in males who often have a long history of smoking and alcohol consumption [8, 3]. To detect laryngeal leukoplakia as early as possible, an examination using video-laryngoscopy and video-stroboscopy is highly recommended. Since the differential diagnosis of a leukoplakia versus carcinoma or chronic hyperplastic laryngitis can often be difficult, to the phoniatrician or ENT-specialist an objetive computer aided ’second opinion’ about the lesions in the video-stroboscopy images might be very useful. 1ENT= ’Ear Nose Throat’

5 Figures and Tables

Cite this paper

@inproceedings{Wittenberg2003AutomaticCO, title={Automatic Classification of Leukoplakia on Vocal Folds Using Color Texture Features}, author={Thomas Wittenberg and Christian Kothe and Christian M{\"{u}nzenmayer and Matthias Grobe and Heiko Volk and Markus M. Hess}, year={2003} }