The Classification Of Rat Neuronal Soma

The human brain is an important area that countless scientists tireless research on it. Before a person is born, the brain has been about 100 billion neurons, about trillions of synaptic connections between neurons, forming a maze of network connections. Because the basic structure of mouse brain is similar with the human brain. So,if we can have a clear understanding of the structure of mouse brain tissue, we will be able to pave the way for carrying out "Human Brain Project". The main object of this study is soma rat neurons in certain brain areas.For researching the neurons, firstly, we should make the data visualization, and search the relationship between the shape and the cell bodies of neurons, making a accurate classify of the cell body using supervision and unsupervised classification methods,through which we can find the abnormal cell body.In this article, firstly, we changed the coordinate of data, draw 528 neurons three-dimensional map, by observing the three-dimensional graphics, classified the soma based on experience. Visualization eigenvalues observe the distribution of data, in preparation for the next step; Secondly, in terms of the classification of neurons, this article selected supervised classification method- tree and unsupervised classification-hierarchical clustering two methods, respectively, do soma data classification. Decision tree belongs to machine learning, C lassify the input class and vector. To avoid over- learning phenomenon, we need to construct a good tree model that can not only fit the training sample, but also classify the unknown sample.Next,under the premise of no prior information, we cut the original data into three categories, and then compare clustering results with the previous classification,and inspect the feasibility of these t hree; Finally,in order to reduce the impact of information overlap, dozens of characteristics has been analysis to extract the two integrated variables,that can represent the eighty percent of the original amount of information. With these two variables integrated cluster again, comparing the clustering results and choose the simplest possible way.After the cell body data visualization, cell shape, data laws are clearly visible. This article divided the soma into more regular spherical(448), the rules of spherical(13), and irregular polygons(67). The Error rate got by supervised and unsupervised methods is less than 0.05, C lassification results are very good, it can pave the way for further research on brain diseases and control.