New Algorithms For Tracking The LV In MRI Sequences

For the clinic-diagnosing disease and cardiac functional evaluation (CFE), it is important to estimate and track the motion of cardiac left ventricle (LV) using Magnetic resonance imaging (MRI) sequences. The motion state of LV can provide the most useful information for the cardiac functional evaluation.To the different problems in practice, three main aspects are involved in this paper: the estimation of motion vector field, the parametric tracking model of LV and probability tracking model of LV. The first aspect is used to construct the spatio-temporal restriction terms for tracking; the other two aspects are the Mulriple Active Contour Model (MACM) and the Generalized Fuzzy Partical Filter (GFPF). Traditionally, the parametric model ACM and probability model PF are respectively used to track an integrated contour and estimate the point-wise motion state and trajectory, but now the GFPF have better flexibility and controllability. The GFPF is not only a simply point-wise tracking method, but its robustion can compromise between the precision and speed on contour tracking.The cardiac motion is non-rigid and there are some ambigious contours and noise in the MR images. Thereby, the current estimation algorithms are basically non-robust. It is very necessary to develop novel methods for obtain more precise estimation results. In this paper, the optimized probability functions and local restriction terms have been presented based on the stochastic, fuzzy and optimal theory framework. In this framework, the Motion Estimation (ME) method, ACM and PF have been improved in such aspects as follows:(1) A Gibbs fuzzy-fusion model (GFFM) is presented based on the Markov-Gibbs model. The results of GFFM are optimal fusion data which partially succeed to that of classical ME algorithms and are improved greatly.(2) Due to the defects of traditional Gradient Vector Flow (GVF), the2 This work was supported by "973" program of China (No. 2002CB312104), Key NNSF of China (No.30130180), and NNSFof China (No.60302022).conception and algorithms of Generalized Fuzzy Gradient Vector Flow (GFVF) are presented. The results of GFGVF are better than the one of GVF.(3) For the traditional ACM used in image segmentation can't be used in motion tracking, the MACM is presented based on the parametric motion estimation algorithm.(4) Using the Sequential Monte Carlo Methods (SMCM) and Generalized Fuzzy theory, the new algorithm of GFPF is presented to track the LV motion. The GFPF can not only be used to describe the motion trajectory, but also be used to track the inner wall boundary of the LV under some appropriate restricting terms. Another, the GFPF can solve the problem of particle degen-erace which always exists in the other methods.In above aspects, the results of GFGVF of the second item provide the important external force for the MACM of the third aspect. The algorithms above have been proved in experiments and can outperform the current popular ones. Furthermore, a set of methods for CFE have been proposed for the clinic-diagnosing.