Abstract: Biological tissues characterization can be approached by non-ionizing optical techniques, in a non-invasive, non-contact way. Optical diagnostic techniques include Optical Coherence Tomography, spectroscopy or fluorescence, among others. Tissue differentiation is difficult to achieve in general with high specificity and sensibility. Spectroscopy is of great interest for this aim, as it provides intrinsic molecular contrast. The different composition and/or structure of biological tissues influence the spectral response. However, the interpretation of spectra is difficult from the raw data, and further data processing is needed. Diffuse Reflectance Spectroscopy (DRS) is particularly well-suited for biomedical applications, as it can work with bulk tissues in reflection, reinforcing the non-invasive character of the technique. DRS has been employed for malignant tissue detection and also for healthy tissue discrimination. These applications require an adequate definition of potential biomarkers for the classification algorithms. The classification process depends strongly on the amount of collected spectra and tissue and specimen variability. In this work several types of ex-vivo porcine tissues are extracted and measured by DRS. Spectral measurements are made on different specimens, and on different points of each sample. Spectra are normalized and several algorithms for dimension and variability reduction are applied, such as Principal Component Analysis or Savitzky-Golay filtering. From these spectra, several biomarkers are proposed for tissue classification, and different classifiers are applied. The results are compared, and the classification efficiency is quantified. The considered approaches could be of particular interest in image-guided surgery or other types of optical biopsy applications.

Authorship

Download reference