The detection of dysplasia and early cancer is important because of the imp
roved survival rates associated with early treatment of cancer. Raman spect
roscopy is sensitive to the changes in molecular composition and molecular
conformation that occur in tissue during carcinogenesis, and recent develop
ments in fiber-optic probe technology enable its application as an in vivo
technique. In this study, the potential of Raman spectroscopy for in vivo c
lassification of normal and dysplastic tissue was investigated. A rat model
was used for this purpose, in which dysplasia in the epithelium of the pal
ate was induced by topical application of the carcinogen 4-nitroquinoline 1
-oxide. High quality in vivo spectra of normal and dysplastic rat palate ti
ssue, obtained using signal integration times of 100 s were used to create
tissue classification models based on multivariate statistical analysis met
hods. These were tested with an independent set of in vivo spectra, obtaine
d using signal collection times of 10 s, The best performing model, in whic
h signal variance due to signal contributions of the palatal bone was elimi
nated, was able to distinguish between normal tissue, low-grade dysplasia,
and high-grade dysplasia/carcinoma in situ with a selectivity of 0.93 and a
sensitivity of 0.78 for detecting low-grade dysplasia and a specificity of
1 and a sensitivity of 1 for detecting high-grade dysplasia/ carcinoma in
situ.