Fast Segmentation Of Hippocampus In MRI Based On3-D Lattice Boltzmann Model

Nowadays, Alzheimer’s disease has become one of the biggest public health prob-lems. Clinically, the extent of atrophy in hippocampal volume is regarded as a diagnosticcriterion for Alzheimer’s disease. It is important to segment hippocampus in MRI accu-rately and measure the volumes. However, it is difcult to realize because the characterof MRI, such as intensity inhomogeneity, low contrast and great noise, combined with thecomplex tissue of brain. In addition, with the fast development of imaging technology,2-Dimage segmentation cannot satisfy the requirment in clinic. Hence,3D image segmentationis coming into ascendance, especially for hippocampal segmentation. It can take full advan-tage of data information acquired from advanced imaging equipment. So research on fastand accurate method of3D image segmentation is desperately needed.From that point of view, this paper analyzes and reviews the current exsiting meth-ods of medical image segmentation, especially3D medical image segmentation. To avoidthe deficiencies of those methods, such as computational complexity, programming com-plexity, low efciency and poor stability, this paper proposes a novel method for3D imagesegmentation based on3D lattice Boltzmann (LB) model, in order to achieve fast hippocam-pal segmentation. A simplified, stable and computational efcient method is able to makeit easy to analog active contour and handle vast amounts of data quickly.This paper proposes a novel lattice Boltzmann model for3D image segmentation,namely3D-LB model, especially designed for segmentation of hippocampus in MRI. Basedon the characteristics of MRI images, the novel3D-LB model is designed with regionalinformation and edge information of data in this paper. The edge information is embeddedin the three-dimensional LB model as the relaxation factor, and the regional information isembedded in the force of LB model.At last,30groups of three-dimensional volume data of hippocampus from ADNIdataset is chosen as experimental data in this paper. The novel3D-LB model is comparedto both3D CV model (Chan-Vese model, the deform model based on regional information)and the standard results on the same volume data. As evaluation criterion for the qualityof image segmentation, symmetric mean absolute surface distance, Hausdorf distance andDice value is chosen to evaluate the result. The experiment result shows that novel3D-LB model get the better performance, and is more efcient than the classical deform model. Thefeasibility on hippocampus segmentation in MRI base on novel3D-LB model is proven.