Accurate Segmentation Algorithm Of Local Cerebral Vessels Based On 3D Stacked Lines And Multi-step Regression

With the development of people’s living standards,cardiovascular and cerebrovascular diseases have seriously threatened human health.Cerebrovascular segmentation also has become a hot issue in the field of medical image follow-up processing.However,due to the complexity of cerebrovascular anatomy,inconsistent gray and contrast in different regions of the brain,and the variability of morbid structures,the precise segmentation of cerebrovascular becomes a particularly difficult problem.Most existing cerebrovascular segmentation algorithms cannot deal with the problem of inconsistent gray and contrast in medical images of the brain,causing these phenomena such as the cerebrovascular fracture and noise points in the segmentation results.We use the context and shape constraints of the cerebrovascular to avoid the influence of noise and repair ruptured vessels with lower contrast.In this paper,based on 3D stacked straight lines,multi-step regression is used to constrain the cerebral blood vessels to obtain a rough segmentation of the cerebral blood vessels.Further,the MAPMRF framework is used to constrain the context of coarse segmentation results of cerebral blood vessels to obtain accurate segmentation of cerebral blood vessels.The main contributions of this article include:(1)3D linear mask structure based on multi-step regression.This method divides the 3D line detection into two 2D line detections.The parameters of the 3D lines are obtained through each linear regression,then using these parameters,the DDA algorithm and mathematical morphology to produce a 3D straight line mask with width and thickness.On this basis,rough segmentation of cerebral blood vessels is performed to reduce the impact of medical image noise on segmentation accuracy.(2)Accurate segmentation of cerebral blood vessels based on the MAP-MRF framework.First use the rough segmentation result to initialize the label field,then use GMM to model the cerebrovascular and background separately,then calculate the energy of the label field and the observation field according to the cerebrovascular and background distribution function and the Potts model,and finally find the minimum posterior energy according to the ICM algorithm to obtain a new label Field until it meets the conditions to stop iteration.The resulting label field is the result of cerebrovascular segmentation.