Research Of Identification Method For Hemocyte On HHT-SVM

Hemocyte count is the main basis of clinical diagnosis to recognize diseases’beingness, types and severity. Currently, most domestic hematology analyzers adoptthe Coulter principle to detect hemocytes. However, the ‘gang’ and ‘edge’ phenomenaappearing in the process of detection make the forms of hemocytes pulsesabnormal.The traditional method of pulses count can’t identify suchabnormal-looking hemocytes pulses, and may cause a leakage metering which willaffect the accuracy of hemocytes count. Therefore, the accuracy of classification andidentification for hemocytes signal has a great importance on its accurate count.Firstly, this paper reviewes the present situation, trend and exsiting problems ofhemocytes analysis at home and abroad, and emphatically analyzes the currentproblems hemocytes count has confronted. And it also respectively expounds theprogress and the present situation of Hilbert-Huang Transform(HHT) and SupportVector Machine(SVM).Secondly, this paper introduces the acquisition principle, the collectiong andpreprocessing of hemocytes pulse signal, and analyzes the cause and influence ofpolymorphous hemocytes signal.Then, this pape respectively expounds the theories of Hilbert-Huang TransformSupport Vector Machine. For Hilbert-Huang Transform, this paper introduces its basicprinciple, two core algorithms, the analysis process and the advantage applied intime-frequency analysis. For Support Vector Machine, this paper summarizes itstherotical basis, classification principles and algorithms as well as the advantages anddisavantages of the classification methods commonly used at present.Aiming at the recognition of polymorphous hemocytes, this paper designs a newalgorithm on the basis of Hilbert-Huang Transform and muti-classification SupportVector Machine after analyzing and comparing these commonly used methods forfeature extraction and classification. According to the time-domain characteristics ofpolymorphous hemocytes, the width, height, peak and valley ratio, gradient ofhemocyte pulses are made as the hemocyte signals’ time-domain features. Hilbert-Huang Transform is applied to extract the spectrum centroids and energycontribution rates of2~5order IMFs as hemocyte signals’ frequency-domain features.The feature vetor of hemocytes is comprised of time and frequency domain features.According to the characteristics’ properties and categories of hemocytes, a muti-classclassifier on the basis of Support Vector Machine is designed and its structure,classification algorithm and other related parameters are determined through thetheoretical analysis and experimental tests. The simulation results on LibSVM haveshowed that this algorithm has a good classification effect on polymorphoushemocyte signals, and the clinical testing results have verified the validity andfeasibility of this algorithm.The main research achievement in this paper is that it has solved the leakagemetering, mis-metering and low recognition precision appearing in the traditionalpulse count based on the research of recognition algorithm for the nonlinear,nonstationary and plymorphous hemotcyte signals. The algorithm designed in thispaper is a beneficial attempt to slove the stability, accuracy and other vital questionsof hemotcytes’ classification and identification, which may help to break out themonopoly of international brand producers of hematology analyzers on the coretechnology of hemocytes identification, and promote the further development of thedomestic hematology analyzers.