Research On Probabilistic Tractography Algorithm In Diffusion MRI

Up to date, the research on nervous system has always been the hot and difficultpoint in the field of neuroscience, it helps us better understand the inner structure ofhuman brains and how brains work. And also in clinical application, it will play a greatrole on diagnosing and treating some neural diseases threatening human’s health nowand in future.Diffusion tensor imaging (DTI) technique is a non-invasive imaging methodbased on magnetic resonance imaging (MRI). By measuring the diffusion tensors ofwater molecules in brain, DTI technique can well tract the base structure of the nervefibers. Until now, DTI is the only technique that can tract the fiber path on an alivehuman, therefore it is very important and realistic necessary to pay more effort toresearching on DTI tractography.According to DTI theory, there is only one diffusion tensor matrix in each voxel,which means there is only one orientation within a voxel. While in crossing area, thereshould be more than one orientation in one voxel. As a result, singe tensor model is notreasonable in crossing area any more, and the tractography based on single tensormodel is incorrect.Broadly, The DTI tractography methods can be classified into two groups:deterministic methods and probabilistic methods. Deterministic technique has beendeveloped for a long time, and it is more mature than probabilistic technique,deterministic methods were widely used in DTI tractography, especially in clinicalapplication. However, deterministic technique has its inherent limitations, such as lowaccuracy in crossing area and bad anti-noise performance. The probabilistic methodsestimate the fiber orientations in a probabilistic way, so they have better anti-noiseperformance, and also probabilistic methods have much higher accuracy in crossingarea than deterministic methods. With these advantages, probabilistic methods are paidmore and more attention to now.In this paper, we put forward a probabilistic tractography method based onparticle filter and two-tensor model, which is used to decompose the single diffusiontensor into to crossing diffusion tensors, which can well describe the crossing structure.Particle filter is a probabilistic filter based on a Bayesian estimator, which has highanti-noise performance. As to the lack of so-called golden standards, we simulated aset of phantoms to test our algorithm. Besides, we collect some real MRI databasesfrom Peaking university Shenzhen hospital to analysis the clinical performance of ouralgorithm. As comparison, a streamline tracking technique (STT) method and Fan Zhang’s particle filter method were used at the meantime. The test results showed thatour method has better performance in crossing area.