Discrimination of moderately differentiated oral squamous cell carcinoma from diabetic oral mucosa using Fourier transform infrared spectroscopy

Oral cancer is a complex malignant head and neck disease which causes uncontrolled cell multiplication. Diabetes is a chronic metabolic condition which is characterized by elevated glucose concentrations. Studies have shown that diabetes is a risk factor for oral cancer, potentially because of the overproduction of reactive oxygen species and the high levels of insulin-like growth factors. The current gold-standard diagnostic methods are time-consuming, invasive, and subject to inter-observer variability. Novel approaches are necessary to identify the disease and malignancies. Fourier transform infrared (FTIR) spectroscopy is a sensitive and reproducible analytical technique based on the detection of the vibrational modes of molecular bonds. This study aimed to use FTIR spectroscopy to identify early biomarkers of oral cancer in patients with diabetes. A total of 10 control (healthy) and 10 oral cancerous (diabetic) samples were used, and 10 independent spectra were acquired from each sample in order to detect the intra-sample heterogeneity. Spectral biomarkers that contributed to spectral variation were identified in the regions of amide I and amide III. Four machine learning models were generated: model 1 between the range of 600 and 4000cm-1, model 2 between 1000 and 1800cm-1, model 3 between 1000 and 1500cm-1, and model 4 between 1800 and 3000cm-1, respectively, and the data were analyzed by principal component analysis PCA, linear discriminant analysis (LDA), a k-nearest neighbors algorithm (k-NN), and a support vector machine (SVM). The prediction accuracy of the models when applying the SVM produced the highest classifier, at >80%. These results support the use of FTIR spectroscopy with machine learning as an adjunct method to the gold standard in the diagnosis of oral cancer.