Using Transformation Knowledge for the Classification of Raman Spectra of Biological Samples

K.-D. Peschke, B. Haasdonk, O. Ronneberger, H. Burkhardt, P. Rösch, M. Harz, and J. Popp (Germany)

Keywords

biomedical computing, tangent distance, prior knowledge, SVM single cell classification, Raman spectra

Abstract

For the classification of biological samples based on Ra man spectra, a robust classifier is necessary. This require ment is met by using Support Vector Machines (SVMs) enhanced by incorporating a-priori knowledge about pat tern variations. In the described approach transformation knowledge is included directly into the classification pro cess by using regularized tangent distance kernels. This approach replaces the standard Euclidean distance in the kernel function by the distance of the linear approxima tion (tangent spaces) of known transformation manifolds. These transformations represent first a global scaling of the spectral values referring to intensity variations, and second a baseline shift by Lagrange polynomials. Experiments are carried out and reported in this paper. The results show, that incorporating a-priori knowledge by tangent distances improves the classification rates substantially, while a lossy baseline correction becomes superfluous.

Important Links:

Go Back