Study On Blood Flow Separation Algorithm Of Diffuse Light Correlation Spectrum Based On Wavelet Packet And ICA

Near-Infrared Diffuse Correlation Spectroscopy(DCS)is a new method for non-invasive measurement of human tissue blood flow.However,due to the structure of human head or skeletal muscle,the blood flow information measured by DCS is the signals mixed with those from superficial tissue and from deep tissues.The purpose of this study is to separate the blood flow information of the superficial and deep tissues from the mixed blood flow signals measured by DCS.When the distance between the light source and detector(S-D)is different,the detected tissue depth is different,and the signal obtained is also different.When the S-D distance is small,the measured data contains more superficial tissue signals.When the S-D distance is large,the measured data contains more deep tissue signals.In this study,a hybrid algorithm of wavelet packet(WP)and independent component analysis(ICA)is proposed to separate and process the two S-D channel of DCS signals,in order to separate the superficial and deep tissue signals.For the DCS measurements,we placed a light source and two detectors on the same side of the human tissue,and the S-D distances are different.The mixed blood flow signals at the small and large distances were measured by DCS respectively.By assuming there is correlation between the superficial tissue and the deep tissue signal,the wavelet packet is used to obtain the subband decomposition of the two mixed blood flow signals.Through taking the mutual information as the standard,the subband with the least correlation is selected from the subband decomposed signals.The selected sub-band with the least correlation(or independent sub-band)is used as the input of ICA,and the de-mixing matrix is obtained to recover the relevant superficial and deep tissue blood flow information.In this paper,the algorithm is initially verified by computer simulation,and then further demonstrated by phantom experiments.Finally,the algorithm is applied to separate the tissue blood flow signals in human experiments.In computer simulations,we assume that the blood flows in superficial and deep tissues are with different waveforms,and we combines the Monte Carlo simulation of photons with the Nth-order linear algorithm to obtain the mixed blood flow signals at the small and large S-D distances respectively.Then the mixed signals are processed with wavelet packets and ICA.The results show that the separated superficial and deep signals are highly consistent with the original assumed signals.In the phantom experiment,the phantom apparatus with injection pump is constructed to mimic the human microvascular blood flow environment,and the DCS instrument is used to collect the signals from two different S-D pairs.The algorithm proposed by us is used to process and analyze the data,and the separated signal is found to be consistent with the true signals generated by the injection pump.For human experiments,the physiological tests of 70-degree table-tilting and cuff-occlusion were carried out,and the blood flow change signals of human tissues at different S-D pairs were measured by the DCS instrument.The mixed signals were processed with the proposed algorithm and compared with the theoretical analysis.In computer simulation,phantom experiment and human experiment,the hybrid algorithm of wavelet packet and ICA has achieved good outcomes of signal separations.This algorithm is able to separate the blood flow information in the superficial and deep tissues,from the DCS signals measured by S-D pairs at different distances,regardless of the tissue structure.It is easy to implement in clinical application and has important practical value.