The Research On The Method Of Brain Mapping Signal Detection

The 21th century will be the brain century, Brain science is playing a more and more important role in exploring the mystery of human brain, developing the neural computer and artificial intelligence etc. Utilizing optical imaging method to research the space-time pattern of brain activities in different stimulus has very important meaning in further investigating the course of brain information processing and the advanced functions of brain.When study the brain function construction, the signals collected by the optical imaging are called intrinsic signals. Detecting the mapping signal(also called mapping pattern), which is activated by the nerve cell activities, from the intrinsic signals is a very important task in studying the brain function construction. The researches indicate that the signal amplitude responded by the nerve cell activities usually only occupies 0.01% of the intrinsic signals'. Such low signal to noise ratio(SNR) brings enormous difficulty for the mapping signal detection. Therefore, with the aid of the modern signal processing technologies, the scholars have proposed a series of detection methods, such as difference images(DI), principal component algorithm(PCA), independent component analysis(ICA), indicator functions(IF), generalized indicator functions(GIF), extended spatial decorrelation(ESD) etc. Based on the predecessors' research work, this article presents a series of new detection methods according to the following two research ways: l)Improve the performance indexes(such as detection performance, robustness and so on) for the existed detection methods with the aid of modern signal processing methods; 2)Combine the research conclusions of other fields(for example blind source separation, clustering, iteration computing for the matrix eigenvalue theory etc) with the specific problems in brain functional optical imaging field. These new methods presented by this article can be divided into two kinds, namely spatial detection methods and temporal detection methods. The spatial detection methods include ICA based on subspace(SBICA), extended spatial decorrelation based-on cycle correlation (ESDBCC) and difference extended spatial decorrelation(DESD). The temporal detection methods include temporal decorrelation source separation(TDSEP) and recursive generalized indicator functions(RGIF).Compared with the existed detection methods, our methods have the following innovations: 1)In order to resolve the main flaw of ICA, namely sources must be independent each other, this article proposes a new SBICA detection method. This method first separates the brain images obtained by the optical imaging into multi subspaces, and then in these subspaces the mapping patterns are extracted from the "noise" patterns using ICA algorithm. Our SBICA method is superior to the ICA method in application range and detection performance in the very low SNR case; 2)From two aspects, namely correlation operation way and the empirical datum structure used by ESD method, two improved ESDBCC and DESD methods are proposed in this article. ESDBCC uses the unidimensional circular correlation operation to replace the original two-dimensional shift correlation operation used by ESD. The simulation results show that the ESDBCC method not only reduces the computational burden sharply, but also has better robustness for parameters setting in noise/noiseless cases; By making use of the characteristic of the orthogonal stimulates experiment, the DESD method respectively gets the two mapping patterns for each experimental conditions, then the two mapping patterns are differentiated to produce the final global mapping pattern. This processing style used by DESD avoids the information loss, that will be encountered when directly differentiating the images of the two conditional experiments as ESD does; 3)To the shortage of requiring strict conditions for the most spatial detection methods' application, this article successfully introduces the TDESP algorithm, which first appeared in the blind source separation field, into the brain functional optical imaging field in the very low SNR case by designing a novel selection strategy for the image pix time series. The theory and simulation results all indicate that the noncorrelation between each image pix time series required by the TDSEP method is more achievable than the independence between each spatial patterns required by the many spatial detection methods in practice; 4)Because the GIF method has the shortages of heavy computation and not suitable for on-line observation, a new RGIF method is presented in this article by combining the GIF method with the iterative computing of matrix eigenvalue problem. The simulation results show that the RGIF method can relieve the computational burden substantially with at almost the same computing precision as that with GIF method.