Implantable Neural Signal Detection And Its Application In Anesthetic Depth Assessment

The implantable neural electrical signals acquisition based on multi-electrode arrays plays anincreasingly important role in the study of neural engineering. It offers a way for researchers toobserve the generation of information flow, synaptic transmission and encoding at the neuronallevel, but there are some urgent problems to be solved in its applications. For example, signals ineach channel of multi-electrode arrays always consist of noise introduced by acquisition system,background noise of neuronal system, and the activities of multiple neurons nearby. Thus, effectivespike signals detection from implantable neural electrical signals mixed with noises and patternclassification to their radioactive sources are the foundation and prerequisite of neuronal encodingmechanics study. For another example, spike rates are not an effective and feasible encodingmethod to express the relevancies between spikes trains and external excitation modes. Therefore,this paper focuses on the detection techniques for implantable neural electrical signals, and appliesthe research results to anesthesia depth assessment to explore the neural coding techniques.The neural electrical activity is an external reflection of modulation of the potassium and sodiummembrane ion channels, and the spikes generated by excited neurons have a certain nonlineartime-varying characteristics. So, linear feature extraction methods such as template matchingmethod and the traditional PCA method may be not suitable in implantable neural signals detection.Based on effectiveness of approximate entropy in describing the nonlinear signal complexity, thispaper proposed a spike feature extraction method using approximate entropy and combined withK-means clustering method to complete the spike pattern classification. Meanwhile, consideredneuronal system is a nonlinear dynamic system, its disorderly spike activities may exist orderedcomponents in high dimensional space. So, this paper proposed another method that using phasespace reconstruction to construct the high-dimensional phase space of spike signals with QRdecomposition method to extract its eigenvalues; then features were further quantified byinformation entropy, and finally blind source separation of spikes was realized by DBSCANclustering method. Since the anesthesia depth assessment is of an outstanding significance in themedical clinic, this paper designed and carried out an animal experiment of implantable nervesignals acquisition under anesthesia. Pulse sequences in certain cortex were used to research theapplication of neural coding in anesthesia depth assessment from firing rate, firing time interval andthe complexity of the firing mode.(1) This paper put forward a spike sorting method based on the approximate entropy. Firstly,multidimensional nonlinear features of spikes were extracted by approximate entropy; then,feature dimension was reduced using KS test to select features with the best separability, and combined with K-means clustering to classify spike signals. For the simulation data and the realexperimental data, the spike pattern classification have achieved better results compared withtraditional methods, our new method for non-homologous spike classification has a certainadvantage.(2) This paper presented an approach based on phase space reconstruction and QRdecomposition of spike feature extraction. For the dynamic performance in spike firing, phase spacereconstruction method was proposed to characterize the dynamic information, and then QRdecomposition was used to extract its eigenvalues in the high-dimensional reconstruction phasespace. Finally, DBSCAN clustering method was used to realize unsupervised classification.Classification accuracy of Simulation data reached98%, and compared with the traditional methodby PCA classification results, it was illustrated that features portrayed by the phase spacereconstruction method combining density clustering can distinguish the difference betweennon-homologous spikes much better. Classification results of experimental data also showed thatour new method can be used as a basis for spike pattern classification.(3) This paper presented a pulse sequence encoding method based on complexity and multi-scaleentropy to make up for the loss of firing information of firing rate and the time interval encoding,and expressed a viable nerve encoding method from the nonlinear characteristics of the dischargemode. In this paper, for clinical application of anesthetic depth assessment, corresponding animalexperiment was designed and carried out to achieve implantable neural electrical activity ofdifferent anesthetic state. Then, the feasibility study of neural coding applications in assessing thedepth of anesthesia was carried on respectively from the average firing rate, three methods based onfiring time interval, Lempel-Ziv complexity and multi-scale entropy discharge mode of neuronsfiring signal. The results showed that the proposed complexity and multi-scale entropy method hada good discrimination in pattern recognition of anesthesia depth, which verified their feasibility as aneural coding approach.