Research On Breast Cancer Diagnosis Based On Pretreatment Of Cells Raman Spectra And Classification Model

In recent years,the application of Raman spectroscopy in disease screening has received increasing attention.Detecting cancer cells based on Raman spectroscopy is helpful for early diagnosis of breast cancer.However,in addition to the necessary biological fingerprint information,Raman spectroscopy also contains many interfering signals,so pre-processing the Raman spectroscopy to reduce noise is a recognized necessary process.However,existing methods for detecting cancer cells using Raman spectroscopy usually perform a certain number of pre-processing in a fixed order first,and then perform pattern recognition on the pre-processed spectrum,without considering the effect of different pre-processing arrangements on subsequent classification effects.And there are many algorithms for different pre-processing to choose from,traversing all algorithm combinations will consume a lot of time and memory resources.To address these problems,this paper firstly studies the pre-processing permutation of Raman spectroscopy,selecting and verifying the top K preprocessing permutations according to the evaluation criteria of the classifiers formed by different permutations,then summarizing the methods of Raman spectra of breast cells based on observation.Next,a selection method of pre-processing algorithm based on cluster search is proposed to solve the selection problem of many algorithms,and the effectiveness,efficiency and the advantages of the method compared with the existing optimization algorithms are proved.This paper studies the effect of pre-processing permutations of Raman spectroscopy on the classification of breast cells.We selected four kinds of pre-processing,such as background calibration,scaling,filtering and squashing to form all possible pre-processing permutations.For each pre-processing permutation,a classifier that recognizes breast cells is trained.Then,the recognition performance of these classifiers is evaluated,and the preprocessing rankings of the top K are screened according to the evaluation scores.And the Mean AUC value of these pre-processing permutations are all above 0.9,and then the effectiveness of the selected pre-processing permutation is verified.We found that the preprocessing permutations of Raman spectroscopy has a great influence on the classification effect of breast cells,and put forward suggestions on the rational selection of the Raman spectrum pre-processing permutations of breast cells.Next,a pre-processing algorithm selection method is designed.For a given pre-processing permutation,the method uses breadth first search tree sorting the nodes of each layer according to the evaluation criteria.A certain number of nodes to expand to the next layer are reserved and remaining are discarded until the last step of the permutation.This method does not need to traverse all combinations of pre-processing algorithms to select the appropriate solution for each pretreatment,so it has significant efficiency.Compared with the full traversal and the existing optimization algorithm,the experimental results fully reflect the effectiveness and irreplaceable advantages of this method.We can select the appropriate algorithm for the pretreatment in each permutation without traversing all the combinations of preprocessing algorithms.We use the top K pre-processed permutations screened by the previous work to conduct experiments.The influence of different values of parameters on the selection result of the pre-processing algorithm is studied,and the appropriate value of the parameter is determined comprehensively by the operation time and evaluation values of the result.Finally,the effectiveness of the method is proved by comparing the mean AUC value of the combination of the preprocessing algorithm selected according to the selection method and the combination with the best effect among all algorithm combinations.Among them,there are five preprocessing permutations directly select the corresponding arrangement.The best preprocessing algorithm combination among all the algorithm combinations for selection has an average AUC of above 0.90 and a maximum of 0.96.AUC values of some combinations are close to the AUC values of the best combination of all algorithm available for selection.And the running time is 1/240 of that in traversing all permutation.And compared with Genetic algorithm and Particle Swarm Optimization in terms of results and running time,it fully reflects the efficiency and effectiveness of the selection method.In summary,based on the research of Raman spectroscopy of five breast cell permutation and selection of pre-processing algorithm combination method,the article puts forward the methods of permutation of Raman spectroscopy selection and efficient selection of pre-processing algorithm combination.It also achieves a great classification of five types of breast cells by Raman spectroscopy,and provides reasonable suggestions for the subsequent pre-processing permutation of breast cells.It is hoped that in the future,the method designed in this paper can be continuously improved and optimized,and it can be used to clinically classify the Raman spectra of breast cells of different molecular types and become an effective auxiliary tool for breast cancer diagnosis.