Detection of Stages of Oral Cavity Lesions Using a Fluorescence Portable Device and Their Classification by Artificial Intelligence Tools.

OBJECTIVE

In the present study, we utilized a fluorescence-based portable device for the detection of oral lesions, as well as AI tools for the classification of the spectroscopic data. The portable device comprises optical components (lens, filters, and beam splitter), a laser, a spectrometer, and other accessories (laptop, fibers, etc.). Laser light (λexc= 405 nm), irradiated onto the lateral border of the tongue (LBT) of the oral cavity tissue, excites the fluorophores FAD (λemi ≈ 500 nm) and Porphyrin (λemi ≈ 634 nm) more significantly than the others.

METHODS

Fluorescence spectra were recorded from three groups, i.e., OSCC, Dysplastic, and Normal, in the range of 450 to 750 nm. Among these groups, OSCC and Dysplastic patients exhibited a significant rise in porphyrin fluorescence.

RESULT: Data analysis was accomplished using AI tools, namely Principal Component Analysis (PCA), K-Nearest Neighbors (KNN), Naïve Bayes, Linear and Quadratic Discriminant Analysis (LDA & QDA), and ROC analysis. Among these AI tools, the PCA-based QDA classifier was capable of discerning among the groups with slightly higher accuracy values (≈ 98%) than the other AI tools.

CONCLUSION

Results reveal that the in-house-built fluorescence device, along with QDA, would be an elegant tool for the identification of oral lesions at the preliminary stage.