Research On Registration Of Multi-b-value Diffusion-weighted Images

Over the recent years,.magnetic.resonance imaging equipment and.imaging technology have developed rapidly.The quantitative analysis of diffusion-weighted images has also been widely used in clinical diagnosis.The prediction of diagnosis,treatment and prognosis can be made by analysis of tissues' diffusion coefficients.However,the movement among diffusion weighted images with different b values can be caused by patients' respiratory and cardiac motion during scan,which will lead to a reduced diagnostic value.Thus the alignment of different images by registration is a precondition for precise quantitative analysis.The change of contrast caused by different degree of intensity attenuation among diffusion-weighted images with various b values and the existence of inhomogeneity in the same image will cause unwanted distortions during the registration process of different b-value images to bo image by conventional registration algorithms,especially for high-b-value images.Freeform deformation model is a commonly used non-rigid registration algorithm,which has been widely used in previous studies.Due to its excellent ability of local motion simulation,it is a very suitable algorithm in medical image registration.However,the free-form deformation model based algorithms are still deficient in registration diffusion-weighted images due to the characteristics of the images,which means the probability of misalignment is greatly increased when registering images with large signal attenuation.To solve this problem,fitting accuracy guided free-form deformation registration was proposed in this paper.Firstly,the signal attenuation information of diffusion-weighted image is first fitted with the model of intro-voxel incoherent motion model and the fitting accuracy is constructed by determination coefficient which is commonly used in statistical analysis to evaluate the fitting property of the model.Fitting accuracy describes the degree of agreement between the signal intensity after fitting and the exponential decay under ideal conditions.A higher degree of agreement indicates that the required deformation is smaller in that voxel,and vice versa.Then,a weight matrix which weights each voxel in the image during registration is constructed by the fitting accuracy to guide the free-from deformation model.Different from the conventional deformation field in free-form deformation which a uniform deformation step is used,the weight matrix can adaptively adjust the deformation step at different positions in the image according to the fitting accuracy on order to obtain the optimal deformation field.The new method combines the fitting accuracy which involves signal attenuation information of images and the free-form deformation which has excellent ability of local motion simulation.It can effectively address the defects of free-form deformation and achieve the precise alignment when registering diffusion-weighted images.In this work,two-dimensional and three-dimensional registration experiments were performed on 11 groups of multi-b abdominal multi-b-value diffusion-weighted images using the free-form deformation based registration algorithm and the proposed method respectively.The Dice ratio was used as the evaluation index of registration results.Compared with the free-form deformation based registration algorithm,the new method obtains higher Dice ratios and the significance of difference was proved by paired t-test.The intuitive and quantitative results show that the proposed method has a better registration performance than the free-form deformation based registration algorithm.