Study On Noninvasive Analysis Of Hematocrit Value By Near Infrared Spectrum

Hematocrit refers to the erythrocytes volume percentage of whole blood, mainlyused in diagnosis of anemia, hemorrhagic diseases, which has the important referencesignificance in clinic. However, the conventional detection methods for hematocrit aretrauma ones which is mainly through vein blood testing blood samples. The methodneeds professional operation with pain, reagent and long detection cycle. Nearinfrared noninvasive biochemical detection has the advantage of no damage, noinfection, simultaneously multi-component analysis, real-time detection and so on. Ithas become one of the hot spots in the present biochemical detection of bloodinternationally.The difficulties of the red blood cell volume near infrared non-invasive detectionmainly faced is weak signal, the organization background interference and detectionwavelength points less leading to difficult to use spectral preprocessing methods. Thispaper use the red blood cell volume of noninvasive detection device to collect humanindex finger tip volume pulse wave signal. The device adopts the16pixels arraydetector, combined with high speed data acquisition system to achieve themulti-channel high SNR signal collection, which solves the problem of weak signaland the less measuring wavelength points. At the same time, the blood volume pulsewave spectral subtraction method processes volume data, to eliminate backgroundinterference from the human skin and muscle tissue. The main research contents and results are:1) We tested the index finger tip of volunteers with different age and different genderusing noninvasive method to acquit the volume pulse wave data. In order to solve theabove problems, the differential spectrum method was introduced;2) We used multi-channel high SNR hematocrit noninvasive near-infrared detectiondevice to collecte forefinger finger volume pulse wave data, which solved the weaksignal and measuring wavelength points less problems;3) We established the hematocrit quantitative calibration model with different datapreprocessing algorithm and modeling method. The optimal hematocrit correctionmodel had been chosen by analysis the forecast accuracy comparison of differentmodels.4) We did further research on BP-ANN and determined the neural network structureoptimization and improved the accuracy of the hematocrit model by adjusting thenumber of hidden layer nodes.5) By experimental verification, the BP-ANN model with first-order differentialpretreatment methods had better forecasting ability, whose correlation coefficient ofcalibration is0.78and the forecast relative standard deviation is7.5%.This paper investigates the error back propagation neural network method andpreferably a hematocrit correction model, which improves the hematocrit modelprecision with noninvasive near infrared spectrum in biochemical detection. It lays thetheoretical and experimental basis for the practical application of noninvasive nearinfrared spectrum in biochemical detection.