PLS-ANN based classification model for oral submucous fibrosis and oral carcinogenesis

Background and Objectives: For effective management of oral neoplasia, autofluorescence, spectroscopy was conducted on patients with different characteristics of oral lesions in vivo. This study tested the possibility of using a multivariate statistical algorithm to differentiate human oral premalignant and malignant lesions from benign lesions or normal oral mucosa. Study Design/Materials and Methods: A fiber optics-based fluorospectrometer was used to measure the autofluorescence spectra from healthy volunteers (NOM) and patients with oral lesions of submucous fibrosis (OSF), epithelial hyperkeratosis (EH), epithelial dysplasia. (ED) , and squamous cell carcinoma (SCC). A partial least-squares and artificial neural network (PLS-ANN) classification algorithm was used to characterize these oral lesions to discriminate premalignant (ED) and malignant (SCC) tissues from "benign" (NOM, OSF, and EH) tissues. Results: The normalized and centerized spectra of the different kinds of samples showed similar but divergent patterns. Our PLS-ANN classification algorithm could differentiate "premalignant and malignant" tissues from " benign" tissues with a sensitivity of 81%, a specificity of 96%, and a positive predictive value of 88%. Conclusions: We conclude that the PLS- ANN classification algorithm based on autofluorescence spectroscopy at 330 -nm excitation is useful for in vivo diagnosis of OSF as well as oral premalignant and malignant lesions