Research On Tumor Detection Algorithm Based On Laser-induced Breakdown Spectroscopy

Tumor detection remains a research hotspot in the biomedical field.The traditional pathological biopsy(gold standard)method has technical bottlenecks such as complicated sample preparation and time-consuming.Imaging detection methods have problems such as low detection accuracy and unobvious tumor characteristics at early stage.Therefore,there was an urgent need for a new method with rapid and high-precision tumor detection.Laser-induced breakdown spectroscopy(LIBS)has many advantages,such as simplicity,minimal loss,simple sample preparation,and simultaneous analysis of multiple elements.It is favored and has been used in metallurgical industry,biomedicine and space exploration.However,in the detection of biological samples,due to the loose nature of the samples and the influence of water,LIBS spectrum fluctuated greatly and spectral feature redundancy,which limited the accuracy of LIBS detection.Therefore,the whole chain of research is carried out around LIBS spectral preprocessing,spectral feature extraction and spectral recognition methods,and rapid tumor detection based on LIBS was realized,laying a solid foundation for its future clinical application.The research results are as follows:(1)In view of the problem of spectral fluctuations caused by spectral scattering in LIBS detection of biological tissues,multiplicative scatter correction(MSC)is used for spectral pretreatment to improve the stability of the spectrum.Taking meat tissue recognition as an example,the k-nearest neighbor(KNN)model was used to classify six types of meat tissue spectra,and the classification accuracy rate was 94.4%.With MSC,the accuracy was increased to 99.1%using the KNN model.The experimental results showed that the MSC has a significant improvement effect on the classification accuracy.(2)Due to the different importance of different spectral lines in LIBS recognition,a random forest(RF)-based spectral feature extraction method was proposed to achieve precise extraction of spectral features.Based on the proposed method,LIBS was used to achieve serum-based nasopharyngeal cancer detection for the first time.RF was used to evaluate the importance of spectral features,and then select the spectral lines whose importance is greater than the average importance of spectral lines.Finally,only three important spectral lines(K I 766.49 nm,K I 769.90 nm and Ca II 373.69 nm)were selected,combined with extreme learning machine(ELM),the recognition accuracy rate was as high as 98.4%.Compared with the principal component analysis feature extraction method,the RF-based spectral feature extraction method requires fewer spectral lines,and at the same time can realize the precise selection of the collected spectral bands.(3)On the basis of spectral feature selection,aiming at the problem of low recognition accuracy of a single classification model,a random subspace method(RSM)was used to construct a multi-classifier decision model.When using a single LIBS spectrum combined with linear discriminant analysis(LDA)and KNN algorithm to identify the four types of tumor serum and health control,the average classification accuracy rates were 88.0%and94.5%,respectively.After using RSM to construct multiple classifiers,the average recognition accuracy of LDA and KNN algorithms can reach 98.4%and 97.2%,respectively,which greatly improves the accuracy and stability of the classification model.(4)Rapid and accurate tumor detection provides scientific guidance for tumor treatment,among which metal anticancer drugs play a huge role in tumor treatment,and the half-life period is a key pharmacokinetic parameter in the development of anticancer drugs.This article proposes a new method for half-life detection of nano hybrid materials(Mn O₂-BSA and Al O(OH)-BSA)using LIBS for the first time.Polynomial fitting and Lorentz fitting were used for spectral pretreatment,and support vector machine regression was used for quantitative analysis.The coefficient of determination was greater than 0.99.Finally,compared with the detection results of inductively coupled plasma mass spectrometry(ICP-MS),the relative errors of Mn and Al quantitative analysis are 5.6%and2.3%respectively.LIBS and ICP-MS have relative errors of less than 5%for Mn O₂-BSA and Al O(OH)-BSA half-lives.The results show that LIBS can be used as a fast and accurate new half-life detection technology.In summary,this paper has carried out systematic research on spectral preprocessing,spectral feature extraction and pattern recognition,which effectively reduced the problems of spectral scattering,spectral feature redundancy and low accuracy rate.Finally,the accuracy of LIBS-based tumor detection and the half-life detection accuracy of nano-hybrid materials are improved.It has laid a good foundation for advancing the application of LIBS for tumor-assisted detection,and provided a new detection method for its pharmacokinetic research.