Detection of precancerous laryngeal leukoplakia using sub-band based cepstral

Precancerous diseases such as leukoplakia, keratosis, Papillomavirus, Erythrolplakia, and other diseases can turn into a cancerous condition represented by cancer of the larynx. However, early detection of these lesions may allow to avoid the occurrence of cancer and provide a possibility for doctors to carry out effective interventions. In this work, a non-invasive automated technique for detecting laryngeal leukoplakia is examined. For reliable detection, we have used perceptual wavelet packet transform (PWPT) to extract the relevant information from the speech signal. In this study, the performance of four Supervised Machine Learning classifiers (SVM, KNN, DT, and NB) is evaluated according to three indicators: accuracy, sensitivity, and specificity. The results obtained in this study are very encouraging. For the classification of female voices, the best performances of the system were obtained with both KNN and SVM classifiers. As for male voices, SVM and DT techniques outperform the rest of the techniques. This study demonstrates that premalignant lesions can be reliably detected, which can aid in the detection of laryngeal cancer at an earlier stage.