Study On Key Technologies For Preprocessing Of Raman Spectrum And Data Analysis Application

As a typical optical detective method,Raman spectroscopy has been widely used in many fields such as biological analysis,disease diagnosis and molecular recognition due to its unique non-invasive,fast,in-situ and extremely high specificity.It has become the most promising optical sensing technology for practical application in the field of biomedicine.However,there are many factors which have a greatly impact on the application of this technology,including weak Raman scattering signal,the difficulty in obtaining high-quality Raman spectrum data,the complex identification of biological samples,the large dependence on analysts for data processing and Analysis,and the low efficiency of processing.Based on the full investigation of previous work,an indepth study on the above issues has been conducted and the following research contents have been mainly completed.1.Aiming at the denoising problem in the preprocessing of Raman spectrum of biological samples,we built a denoising model based on the Back propagation network considering the characteristics of the data.In addition,this method has been compared with the Fourier transform method,Wavelet threshold method,sliding window average method and the Savitzky-Golay filter method.This method avoids complex parameter optimization settings,and at the same time obtains the denoising effect that is almost the same as the optimal wavelet transform method.It greatly simplifies the parameter setting and further improves the efficiency of the preprocessing of Raman spectrum.2.In view of Raman spectrum baseline drift caused by the fluorescence interference from the complex biological samples,we proposed a simplified Lorentz peak function model and a new baseline calibration method in combination with the least square method.The calibration performance of new method has been tested by the simulation for Raman spectrum with four different baselines including Sigmod,Gaussian,Exponential and Ploynomial.Finally,a method of preprocessing of Raman spectrum,combining denoising with baseline calibration,was proposed and verified by simulation data and spectral data of experimental biological sample.3.The identification of spectral peaks is the key to qualitative and quantitative analysis of Raman spectrum data of biological samples.Here,a new peak recognition method based on deep learning framework is proposed.The identification test in a large number of simulation data and RRUFF public experimental data indicate that,the proposed method has relatively small error and high recognition accuracy compared with the traditional continuous wavelet transform and the bi-scale correlation algorithm.It has obvious advantages in identifying weak peaks and overlapping peaks,and the efficiency is the highest under the same hardware configuration.At the same time,compared with other methods,it can effectively identify the cosmic spikes and simultaneously complete the identification of the peak width.All of those would provide a reliable method of data processing for qualitative and quantitative data analysis,reduce the cost of long-term manual comparison and analysis and improve the overall efficiency of data processing.4.Due to the complex components of tumor samples,it is difficult to get the differential fingerprint features because of the Raman spectra of normal and abnormal samples are extremely similar.A total of 28750 spectral data of normal tissue and gastric cancer tissue samples,and 24000 spectral data of melanoma samples and normal samples were collected.After preprocessing those data including denoising,background baseline calibration,we choose the valid data to establish a classification network and a feature extraction model.Through the Raman imaging based on extracted feature information,we found the characteristics extracted which can not only reveal the difference in cell contour and cytoplasmic ratio distribution between normal skin and melanoma,but also identify the differences for the prognosis of gastric patients in different stages.These differences are the same as the HE staining results of the corresponding samples.This further verify that machine learning methods can play an important role in Cancer detection based on Raman spectroscopy technology,greatly improve the efficiency and accuracy of manual analysis.All of the above methods have actively explored the application of Raman spectroscopy for early Cancer screening and diagnosis,which is helpful to promote the further development and clinical application of the Raman Spectroscopy.