Towards a computer tool for automatic detection of laryngeal cancer

Abstract

Vocal pathologies are most of the time due to the irregular behavior of the vocal cords during phonation. This irregular vibration can be induced by benign or malign tumor on the cords and should result in irregular values of used parameters. In this paper, we present a new tool for automatic detection of pathological voice particularly laryngeal cancer. A preprocessing approach was based on the glottal descriptors extracted from the air flow resulting of the vocal cords called Glottal Wave Flow (GWF) or Glottal Source. The latter was analyzed in order to classify it according to its class (healthy/ cancer). Neural networks are used for this classification. In fact, the originality of this work lies essentially upon the proposition of new selection methods based on statistics measures for choosing the glottal parameters which provide a better discrimination voice. Experimental results show the ability of the proposed measures to discriminate the pathological voice from the healthy voice and to achieve a very good identification rate of laryngeal cancer.