No abstract

Objective To test the hypothesis that Fourier transform infrared (FTIR) spectral imaging, coupled with multivariate data processing techniques, can image the spatial distribution of matrix constituents in native and engineered cartilage samples. Methods Tissue sections from native and trypsin‐digested bovine nasal cartilage (BNC) and from engineered cartilage, generated by chick sternal chondrocytes grown in a hollow fiber bioreactor, were placed either on calcium fluoride windows for FTIR analysis or gelatinized microscope slides for histologic analysis. Based on the assumption that cartilage is predominantly chondroitin sulfate (CS) and type II collagen, chemical images were extracted from FTIR spectral imaging data sets using 2 multivariate methods: the Euclidean distance algorithm and a least‐squares approach. Results Least‐squares analysis of the FTIR data of native BNC yielded a collagen content of 54 ± 13% and a CS content of 37 ± 16% (mean ± SD). Euclidean distance analysis of measurements made on trypsin‐digested BNC demonstrated only trace amounts of CS. For engineered cartilage, the CS content was significantly lower (15 ± 5%), while the collagen content (73 ± 6%) was significantly higher than biochemically determined values (CS 34%, collagen 5%, protein 61%). These differences are due to the fact that the dimethylmethylene blue assay overestimated the CS content of the tissue because it is not specific for CS, while the FTIR spectral imaging technique overestimated the collagen content because it lacks specificity for different proteins. Conclusion FTIR spectral imaging combines histology‐like spatial localization with the quantitative capability of bulk chemical analysis. For molecules with a unique spectral signature, such as CS, the FTIR technique coupled with multivariate analysis can define a unique spatial distribution. However, for some applications, the lack of specificity of this technique for different types of proteins may be a limitation.

A combination of Fourier transform infrared (FT-IR) spectroscopy and microscopy, FT-IR microspectroscopy, has been used to characterize sections of human colorectal adenocarcinoma. In this report, a database of 2601 high quality FT-IR point spectra from 26 patient samples and seven different histological structures was recorded and analyzed. The computer-based analysis of the IR spectra was carried out in four steps: (1) an initial test for spectral quality, (2) data pre-processing, (3) data reduction and feature selection, and (4) classification of the tissue spectra by multivariate pattern recognition techniques such as hierarchical clustering and artificial neural network analysis. Furthermore, an example of how spectral databases can be utilized to reassemble false color images of tissue samples is presented. The overall classification accuracy attained by optimized artificial neural networks reached 95%, highlighting the great potential of FT-IR microspectroscopy as a potentially valuable, reagent-free technique for the characterization of tissue specimens. However, technical improvements and the compilation of validated spectral databases are essential prerequisites to make the infrared technique applicable to routine and experimental clinical analysis.