Spectral Quantitative Analysis Of Blood Glucose,Total Cholesterol And Triglyceride In Whole Blood

Hyperglycemia,cholesterol and triglyceride are the main diagnostic indicators of diabetes and cardiovascular disease.Therefore,it has an important significance to explore the low-cost and reagent-free quantitative analysis method of hyperglycemia,cholesterol and triglyceride based on whole blood spectra for rapid diseases screening.However,due to the high complexity and low concentration level of some analytes in whole blood samples severely limit its development.Special collection of medical samples,rare diseases in a population,imbalance information distribution often result in highly imbalanced proportions of normal and abnormal instances.Small samples and information imbalance will seriously affect the performance of the prediction model.Hence,the critical problem of building a stable prediction model is to expand the number of minority samples and improve samples information space.In this paper,we firstly studied the generation,optimization and application of virtual samples for the minority of whole blood spectra samples.Secondly,we established a reliable model to effectively predict the concentrations of blood glucose,cholesterol and triglycerides by combining virtual sample construction technology and partial least squares regression(PLSR).Finally,the performance of other machine learning modeling methods for the quantitative analysis of blood spectral was analyzed and compared.The research contents of this paper are as follows:(1)To solve the problem of small samples in whole blood spectra,a novel virtual sample generation(VSG)method based on MD-MTD(Multi-distribution Mega Trend Diffusion)was proposed,termed Hybrid-MTD,which further expends the training sample set and improves the information distribution of the samples space.(2)In order to research the boundary setting problem of virtual samples and its influence for model performance,the Particle Swarm-Hybrid Mega Trend Diffusion technique(PSO-HyMTD)was proposed.The upper and lower boundaries of virtual samples can be set more reasonably through PSO-HyMTD.(3)To evaluate the performance of different algorithms in the quantitative analysis of whole blood samples,Partial least squares(PLS),interval interval partial least squares(iPLS),stack partial least square(SPLS)and extreme learning machine combined with particle swarm optimization(particle swarmoptimization-extreme learning machine PSO-ELM)for the quantitative analysis of glucose,total cholesterol and triglyceride based on whole blood spectra were performed and analyzed.