Research On Improved Level Set Segmentation Algorithms Of Infant Brain MR Images

The brain is a very important part for human.Infancy is a critical period of brain development.In this period the brain is developed rapidly and haves strong plasticity,while the probability of suffering from a variety of encephalopathy is also higher than in other periods.The exploration of infant brain disease early diagnosis method has great significance.Magnetic resonance image has the higher spatial resolution and won’t cause ionizing radiation to human bodies.So the magnetic resonance image is the first choice for the image technique of brain structure lesions in infant brain.The infant brain MR images have some features while are not conducive to segmentation.The first one is the intensity inhomogeneity,which is caused by the inhomogeneity of the RF field.The second one is the partial volume effect,which can lead to brain tissue boundary discontinuity,unclear and relatively fuzzy.The third one is the effect of noise,which is caused by the influence of the physical reason.The fourth one is the phenomenon of gray matter and white matter reversion during the process of the brain development.These characteristics determine the research of the segmentation algorithm for infant brain MR images are meaningful and challenging.This thesis mainly research the level set aiming at the characteristics of the infant brain MR image.The main work of this thesis lists as follows:(1)According to the noise of the infant brain MR image,the new model is proposed based on the study of the CV model and the LBF model.The new model uses the average intensity of local regions in the LBF model instead of the single intensity value in the CV model to calculate the level set energy function.The new proposed model,which is added the average of local regions,can have a good segmentation for the noisy image.Experiments show that the improved algorithm can segment the noisy images well.(2)According to the low contrast of the gray and white matter and intensity inhomogeneity of the infant brain MR image,the improved model is proposed,which is combining the information of the global region,the information of the local region with the fuzzy logic.Adding the fuzzy logic to the proposed model can improve the robustness of the parameter selection and the initial curve selection.Adding the information of the global and local region to the proposed model can segment well the brain images with the low contrast of the gray and white matter and intensity inhomogeneity.Experiments show that the improved algorithm can segment the infant brain MR images well.The initial curve haves little effect on the segmentation.And the parameter selection is robust.(3)According to the blurred boundaries of the infant brain MR image,the new model is combining the local Gaussian distributing fitting model with the prior information.The prior information of the image is added to the local Gaussian distributing fitting model.The algorithm analyzes the gray distribution of each pixel in the image.Then the prior information of the image is introduced.Next local Gaussian distributing fitting energy is defined by using the kernel functions to.Last the fitting energy of the whole image is obtained by further integration.Qualitative and quantitative experiments show that the improved model can segment the infant brain MR images well.