A Novel Algorithm For White Matter Fiber Reconstrction Based On Higher Order Tensor

Accurate fiber tracking promises to have a high impact in fundamental neuroscience and its clinical applications. Global fiber reconstruction aims at finding the best fiber configuration that describes the measured data based on complex minimal path method on whole volume of brain voxels. With the increasing of the resolution of acquisition diffusion data, traditional tensor models and algorithms cannot meet the demand of reconstructing fiber accurately.In the model estimates, the existing high-order tensor imaging model solves the second-order tensor model (DTI) is difficult to characterize the complex fiber structure of the problem, but there are characteristic direction of fiber extraction complex, computationally intensive and low angular resolution and other issues. This paper proposes an improve method for the tensor modle. The details are listed as follows:On the basis of higher order tensor model, the paper presents a novel method of eigenvector extraction using subdivision theory and iterative search. According to the characteristics of the tensor model, first using the grid subdivision, we quickly determined the direction of the regional characteristics roughly; further according to the area segmentation and iterative we obtain the accurate of high order tensor model features direction. The method solved the existing symbols calculation method is easy to fall into the local extreme value point or the search errors caused by no convergence and computational efficiency. Using synthetic data, we analyzed the recognition ability and the precision of Eigen direction in different numbers and different point of fiber cross cases. Compared with the existing symbols calculation method, stable fiber characteristic direction can be obtained by our algorithm under6th and higher order model.The existing high-order tensor model describes methods and characteristics semi-definite bring large number of calculation and discrete computational errors. This paper intends to establish a model of arbitrary order tensor polynomial theory through the square and to ensure non-negative characteristics, and to establish a continuous high-order tensor fiber orientation distribution function of the convolution model with pulse basis functions, through an iterative deconvolution algorithm is highly accurate and stable fiber orientation distribution function.synthetic and real data experimental results show that he deconvolution spherical model based on higher order tensor proposed in this paper, increasing the resolution of the original high-order tensor model to more accurately reflect the characteristics of the complex fiber itself, which is more reliable for the further study of fiber beam tracking algorithm to create a good condition. Subsequent studies can follow this algorithm is applied to a variety of fiber bundle tracking algorithm, better white matter fiber tracking results, and thus better used in clinical studies.