Characteristic wavenumbers of Raman spectra reveal the molecular mechanisms of oral leukoplakia and can help to improve the performance of diagnostic models

Abstract

The correct diagnosis and the prompt treatment of oral leukoplakia (OLK) can efficiently prevent OLK from undergoing malignant transformation to oral squamous cell carcinoma (OSCC). However, the diagnostic model for distinguishing normal mucosa from low-grade dysplasia, as well as high-grade dysplasia from OSCC was not well established in a previous study. In this study, the characteristic wavenumbers in the Raman spectra were first identified by the variable selection methods. Then, the intensities at these wavenumbers were used to classify the biopsies. As a result, the accuracies achieved by the intensities at the characteristic wavenumbers were 70.5% and 94.0% for the classification of normal vs. low-grade dysplasia and high-grade dysplasia vs. OSCC, respectively, which were greater than those (accuracy of 65.4% and 88.0%, respectively) using all the intensities in the Raman spectra. Our results suggested that constructing a diagnostic model with the intensities at the characteristic wavenumbers can improve the identification of the different lesions of oral mucosa. Moreover, most of the Raman intensities for predicting normal vs. low-grade dysplasia indicated that the transformation from normal mucosa to low-grade dysplasia was associated with the changes in the contents of lipids, while most of the intensities for predicting high-grade dysplasia vs. OSCC indicated that the transformation from high-grade dysplasia to OSCC was associated with changes in the contents of proteins and nucleic acids. Our findings can be helpful for diagnosing the various grades of OLK with dysplasia and understanding the molecular mechanisms of potential malignant transformation of oral leukoplakia.