A Study Of Feature Classification Algorithm For Multisample Data In Low-cost Medical Instruments

The low-cost health care is one developing direction of current medical, which issignificant for the construction of rural health care system. Low-cost medicalinstruments need to achieve the basic function of the instruments under conditions thatreduce the manufacturing and operating cost as much as possible. Improvement of thealgorithms for instrument can promote the reduction of costs, and use minimalhardware resources and energy to achieve much more functionality. In this goal, we, byusing a variety of mathematical methods, have improved core algorithms of the ECGworkstation and blood cell analyzer on the work of the previous.Firstly, wavelet analysis theory and support vector machine methods are used toachieve the design of core algorithms for ECG workstation. ECG filtering process usesbiorthogonal filtering algorithms, removing undesired signal and paving the way forthe subsequent processing. Decomposition of ECG signal is implemented to four scaleusing discrete wavelet analysis method, and the data reconstructed from the four-scaledecomposition. Then, QRS wave group, T-wave and P-wave of ECG are also identifiedby wavelet analysis theory combined with Mallat algorithm. In addition, with theimplementation of support vector machine algorithm, the ECG workstation is capableof learning by itself, and able to provide personalized diagnosis for different patients.The experiments show that this method can achieve a fast and accurate analysis of theECG signal and provide real-time information to medical staff for medical diagnosis. Italso greatly reduces their workload at the same time.Secondly, the design of core algorithm for blood cell analyzer is chiefly realizedby mathematical morphology algorithm. In consideration of the characteristics ofsignal, the Butterworth filter is implemented, and achieves a desired filtering effect. Inorder to get accurate information for blood cell populations, we use mathematicalmorphology methods to identify blood cells pulse, namely identifying the starting point,end point and peak point for pulse. The pulse information will be screened to get validpulse data, and then achieve three categories of leukocyte group by the floatinglandmarks and ultimately calculate the various parameters of the blood cell groupsaccording to the classification results. At the same time, the blood cell populationdistribution histograms are fitted by the nonlinear least squares method and the LMalgorithm. The fit result is more convenient to display and storage, improving theefficiency of the instrument. The experiments and analysis above show that these algorithms are able to accurately give various parameters of blood cell group ofpatients, and to provide a comprehensive reference to the medical staff for thediagnosis.ECG waveform detection and classification of ECG have obtained good results.We have finished the detection of QRS, P and T wave. In the ECG classification area,we use support vector machine method combined with doctors diagnosed results, sothat the classifier can form a self-learning function of the classifier, the effect of thisclassifier is more accurate. In this paper, we use three classification algorithms in bloodcells instrument, and obtain the blood cell histogram, and finally calculate theparameters required for clinical diagnosis, which can provide a reference basis fordisease. The recognition of P, T wave waveform is very difficult. We have to achieve amore accurate algorithm for its detection in future research areas.