Classification of the Oral Mucosal Lesions Using the Principle Component Analysis Based Naïve Bayes Classifier: Fluorescence Spectroscopy as a Detection Device

Detection of mouth cancer also referred as oral mucosal lesions has been accomplished by adopting fluorescence based compact device and the classification of the lesions are being executed by the machine learning methods. In the machine learning method, principle component analysis (PCA), naive Bayes classifier, and ROC analysis are utilized. PCA is a dimension reduction method and employed to reduce the dimension of fluorescence data. Naive Bayes is a probabilistic model which is based on the Bayes theorem is utilized for data classification. However, ROC is implemented for the calculation of sensitivity, accuracy, specificity, and AUC values. Fluorescence data was recorded from the lateral boarder of tongue (LBT) of the oral cavity. Using the device, in-vivo measurements were carried out on 20 OSCC patients (55 LBT sites), 18 dysplastic patients (49 LBT sites), and 36 healthy volunteers (60 LBT sites). In the spectra, presence of two fluorophores namely FAD <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(k_{\text{emi}}\approx 500\ \text{nm})$</tex> and porphyrin <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(\lambda_{\mathrm{e}\mathrm{m}\mathrm{i}}\approx 634\ \text{nm})$</tex> were seen. We observed an enhancement in the porphyrin fluorescence with the evolution of cancer. Data analysis carried out by the naive Bayes classifier was proficient to differentiate among Normal to OSCC, Normal to Dysplasia, and Dysplasia to OSCC with the sensitivity of 82.35%, 81.25%, 82.35% and specificity of 94.44%, 100%, and 93.33% respectively. Results indicate that fluorescence spectroscopy along with the PCA based naive Bayes classifier could be an acceptable motoring tool for the identification of the oral lesions.