Estimation Model And Algorithm For Brain Fiber Orientation Distribution Under Partial Volume Effect

The fiber imaging of brain white matter based on diffusion magnetic resonance imaging is the only non-invasive way to explore the microstructure of brain tissue.In a variety of diffusion MR methods,the high angle resolution diffusion imaging(HARDI)technologies not only overcome the shortcomings of diffusion tensor imaging in reconstruction of crossing fiber but also require less diffusion samples,and become a hotspot in brain fiber imaging.The spherical deconvolution model is one of the most efficient methods based on HARDI,and greatly improves the accuracy of fiber reconstruction.However,due to the relatively large volume of sampled voxels,there are non-negligible partial volume effects between brain tissues,which will lead to inaccuracy of traditional fiber imaging methods.Aiming at the above problem,we design a new model which is based on the traditional spherical deconvolution,and solve the high harmonic truncation problem of the spherical harmonic function in the spherical deconvolution model and the ill-posed least squares method.The main works of this paper are as follows:(1)Considering the partial volume effect of brain,we propose a multi-responses kernel function model.This model is used to obtain a new fiber direction distribution function based on the isotropic and anisotropic diffusion signal of brain tissues.Thus the isotropic signal can be separated from the diffusion weighted signal,which overcomes the partial volume effects of the brain tissue and provides a more accurate estimation of fiber orientation.(2)We use dictionary basis function to represent the fiber orientation distribution function,and propose a new new quantitative index to measure the strength of the isotropic signal.(3)We propose a Richardson-Lucy(RL)algorithm to solve spherical deconvolution model with multi-responses kernel function.The new algorithm use the additional prior knowledge to solve the high degree of ill-posed least squares algorithm by adding the total variation regularization and sparse regularization term to the model.The experiments of simulated data and actual human brain data show that our method can get more accurate fiber directions and improve the fiber angle resolution significantly.The newquantitative index is more effective compared with the traditional quantitative indexes FA and GFA.