RESUMEN
We studied colon carcinogenesis using Fourier-transform infrared (FT-IR) microspectroscopy, an evolving method that allows the nondestructive assessment of the chemical composition of cells and tissues and of the in situ relationship between molecules, and assessed its diagnostic potential. Mid-FT-IR spectra were obtained from frozen colon tissue samples of normal (C57BL/6J) and Min (Apc(Min) mutant) mice, the latter recapitulating key features of human colon carcinogenesis. Classic spectroscopic analysis demonstrated marked differences in the Mid-FT-IR spectra between normal and dysplastic tissues, especially regarding peak positions and band intensity ratios in the regions 1800 to 985 cm(-1) and 3000 to 2700 cm(-1), reflecting changes in cellular nucleic acids, phosphates, and carbohydrates. Analysis of the spectra using the multivariate methods backpropagation neural networks, decision tree, adaboost with decision tree, and support vector machine, which interrogated the intrinsic dimensionality of IR spectra, revealed that their sensitivity was between 91.1% and 100% and their specificity between 94.1% and 100%, with the outcomes of the Support Vector Machine algorithm being identical to those of histologic analysis. FT-IR microspectroscopy holds great promise not only as a method of ascertaining changes in the chemistry of the neoplastic cells but also as a diagnostic tool, especially for early stages of carcinogenesis not detectable by other means.