Research On Traumatic Brain Injury Lesion Segmentation

Digital image segmentation is the basis of image processing; and it significantly affects the correctness of image processing. Active contour model of digital image segmentation has been studied by many scholars; because of rigorous mathematical theory and the dynamic evolution of the contour curve takes the advantage of solving image segmentation which other segmentation algorithms cannot.This thesis mainly studies the segmentation algorithm of brain injury lesions. According to the advantages and disadvantages of active contour model, some improvements and innovations are put forward in order to quickly and accurately segment the brain injury lesions. The specific work arrangement is arranged as follows:1. It is know that the global region based-on active contour model can segment images with weak boundaries, while it is unstable to segment the inhomogeneous image; and the local region based-on active contour model can segment the inhomogeneous images very well, while the efficiency is not high. Through an energy function of the combination of the information of global and local region to segment the inhomogeneous image. In order to control the weight of the global and local region in the energy function and to further promote the speed of image segmentation, this thesis introduces an adaptive weight, which can reduce the computational complexity of the energy function. This algorithm is named adaptive weight hybrid active contour model. The experiments show that the proposed algorithm can extract the lesion area quickly and accurately. However, this method requires rigorous initialization.2. Since the adaptive weight hybrid active contour model requires correct the target area, this article brings forward fuzzy adaptive hybrid active contour model segmentation to extract the brain injury. Firstly, the improved clustering algorithm is utilized for medical image clustering segmentation; and the clustering segmentation result can then be used as the initial target of adaptive hybrid active contour model.The experiments show that it cannot only improve the segmentation speed of lesion area, but also the accuracy of lesion area segmentation.