Research On Human Cardiac Fiber Reconstruction Based On Particle Filtering In MRI

Medical research shows that many cardiovascular diseases can cause myocardial fibrosis. In order to obtain structural information of human cardiac fibers, some tracking methods have appeared, such as biopsy, staining, and chemical tracking. With the development of magnetic resonance imaging technology, cardiac fiber tracking based on magnetic resonance diffusion imaging has attracted more researcher’s interest due to its non-invasive characteristics. This research about cardiac fiber tracking method has important theoretical and clinical value for further exploring pathological changes of cardiac muscle fibers and revealing its causes.Fiber tracking based on magnetic resonance diffusion imaging estimates the fiber’s orientation by analyzing the signal of water diffusion. However, due to the effects of tissue structure and image noise, much more instability can not be avoided in the fiber tracking process. Therefore, compared with the deterministic method, fiber tracking based on probability is more conducive to improve the accuracy, especially in the condition of low angular resolution. How to improve the quality of fiber tracking using probabilistic method has become a hot issue in this field. In order to describe the posterior probability distribution of fiber’s orientation, particle filtering is introduced in our work. Some crictical problems have not been studied sufficiently, such as observation equation, the design of importance density function and adaptive strategies. Based on the above study motivations, this work pay more attention on the listed contents:(1) For the problem of inefficient calculation and convergence in traditional particle filtering, adaptive particle filtering based on anisotropic diffusion is proposed to reconstruct the cardiac fiber path. The proposed method dynamically adjusts the intensity of the disturbance and the number of particles in the prediction stage reference to the diffusion fractional anisotropy value at each voxel. In this framework, time cosumed is reduced and a better tracking result can be obtained. Firstly, at the starting voxel of the fiber path, initialization is taken as the mean of six neighborhoods’ orientations, which is used to ensure particle filtering to converge as soon as possible; then, in the course of fiber evolution, particle number is varied according to the current diffusion status. Parameter in important density function is adjusted according to FA value. So a flexible particle set can be sampled. Experiments results showed that: the proposed method has a strong anti-noise ability in cardiac fiber tracking. System computational cost dropped by an average of 50%, significantly improve system efficiency.(2) For the problem of a big error existing in the existing models, a new observation model is proposed based on tensor rotation invariant properties. First ly, fit DWI data to calculate the diffusion tensor for each voxel; then, construct a rotation matrix according to the position relationship between a given potential fiber direction and the primary diffusion eigenvector; finally, get a new observation model by keeping the tensor shape invariant in the rotation operation. Compared with the constraint model and compartment model, observation model proposed in this paper can be more effectively describing the state of diffusion of water molecules in the tissue. Model error has been reduced using this novel observation model. Experiment results show that the observation model based on tensor rotation invariant features can effectively improve the particle quality in importance sampling, more accurate posterior distribution of the fiber’s orientation has been rebuilded, the accuracy of the reconstructed cardiac fibers has also been significantly improved.(3) For the problem of a big error existed in region of fiber crossing and border under low angular resolution condition, a correction method is proposed to improve the quality of cardiac fiber reconstruction. Firstly, the region is segmented by analysis of the anisotropic diffusion threshold value for fiber crossing or boundry fiber, in which centrifugal diffusion may be existed; then, a correction operation is taken in the evolution of the fiber to improve the reconstruction accuracy. In the region of fiber crossing a larger evolutionary step is used to reach across the region to maintain the fiber’s direction of travel purposes; for the region that a boundary diffusion centrifugal fiber existed, a angle deflection correction is taken to inhibit the fiber deviation caused by centrifugal diffusion phenomena. Experiment results on synthetic data and real cardiac data show that: correction strategy in the fiber econstruction can reduce the error caused by partial volume in fiber crossing and boundary centrifuge under low angular resolution.In addition, in order to make cardiac fibers reconstruction method better meet the needs of practical applications, some research has also been carried out on parallel accelerated method based on the Open MP multicore env ironment particle filter. Firstly, parallel analysis has been make on each particle filtering unit in the cardiac fiber reconstruction; then, a design has been provided for thread management and memory management; and finally, some details about parallel peocess was discussed for the reconstruction of the cardiac fibers. Experiment results showed that: through parallel computing to accelerate its efficiency for particle filtering in the cardiac fibers reconstruction, a significant improvement can be reached.In summary, the research work of this paper is to improve the efficiency and accuracy of cardiac fiber recontruction based on particle filter. It is an important reference for other tracking method based on particle filtering. Our algorithm is implemented at a low angular resolution and single-order tensor conditions and is easy to migrate to higher-order tensor system and the Q-ball and other complex diffusion model to further improve the performance of complex fiber reconstruction.