Conventional Machine Learning Techniques with Features Engineering for Preventive Larynx Cancer Detection

Abstract

Larynx cancer is developed from precancerous state. Some precancerous lesions such as Keratosis, Leukoplakia, Ery-throlplakia, Papiloma virus, etc., can be transformed into a cancer if they are note treated in time. In this paper, we propose a non-intrusive technique to detect precancerous lesions at an earlier stage. Hence, these lesions can be treated as soon as possible. The idea is based on the analysis of the human voice in order to detect pertinent acoustic features able to discriminate pathological voices with precancerous lesions from normal ones. We have tested a large number of speech acoustic features. A feature engineering methodology leads us to choose the most pertinent features. To detect mentioned lesions, several classification techniques are tested. Experimental results show the validity of the idea.