A multivariate statistical algorithm for analyzing fluorescence spectroscopy of oral squamous cell carcinoma-an animal model approach

A multivariate statistical algorithm has been developed to evaluate its diagnostic ability on oral squamous cell carcinoma by using a 7,12-dimethylbenz[a]anthracene (DMBA)-induced hamster buccal pouch carcinogenesis model. The hamsters were divided into a calibration set and a prediction set comprising twenty four animals each. In each set, the animals were categorized into four type groups according to the duration of DMBA application. The partial least square (PLS) analysis were used to dimensionally reduce the input variables and extract the useful diagnostic information from the original data of fluorescence spectra. The logistic regression method, adopting the first three factors obtained from PLS, established a probability-based algorithm for discriminating the samples of hamsters with different cancer transformation stages. By combining the two methods, the correctness of classification in calibration and prediction set were 96% and 88%, respectively