Raman spectra calibration, extraction and neural network based training for sample identification

This research is concerned with a series of preliminary work on a complete procedure for Raman spectra identification of sample tissues to associate the Raman spectra with the medical diagnosis. It consists of the spectrum capturing using Raman spectrometer, spectrum calibration, background spectrum extraction as well as the neural network based data training and identification. Raman spectroscopy technology is able to provide the quantitative morphological and chemical information about tissue compositions. By the principal component analysis and control oriented identification, different samples can be distinguished on a basis of the Raman spectra. Least squares estimation is applied to extract the intrinsic Raman spectrum and then back propagation is used for neural network Raman sample data training. Numerical simulations are conducted on the whole process with the normal tissue samples. The future research will be based on the abnormal tissues, so that we will formulate an easy systematic approach to distinguish between normal tissues and abnormal tissues. The long-term objective is to create a real-time approach of the tissue sample analysis using a Raman spectrometer directly mounted at the end-effector of a certain medical robot, which is helpful to enhance the remote controlled robot surgery.