| Vascular segmentation of brain CTA images is very important in the clinical diagnosis of cerebrovascular diseases,and it is also a challenging task.It is of great theoretical significance and application prospects to study the technology of blood vessel segmentation in 3D brain CTA images based on deep learning.Aiming at the characteristics of brain CTA images,based on the full convolution neural network V-Net in deep learning method,Vessel-VNet,which is suitable for cerebral vascular segmentation network,is proposed.Firstly,in view of the uncertain brain position of CTA images in different cases,the poor definition of fine vessels and the interference of other tissues in the image,a data preprocessing method for blood vessel segmentation of brain CTA images is proposed,which mainly includes brain localization method to extract the location of region of interest in the brain and the method of limit the range of CT values to exclude the interference of unrelated tissues.Secondly,for the characteristics of different blood vessel thickness and most thin,the appropriate times of down-sampling is used in the network to take into account the segmentation of thick and fine blood vessels.Then,considering the details of small blood vessel segmentation,the dilated convolution is applied to the cerebrovascular segmentation network to increase the output receptive field to extract more complete vascular context information at high resolution.Finally,a new loss value calculation method is designed in the network to solve the problem of insufficient feature fusion in the process of up-sampling.The effectiveness of the proposed Vessel-VNet network is verified by testing 20 sets of randomly selected brain CTA data provided by project partners.The experimental results show that the new Vessel-VNet network achieves better performance,with an average Dice coefficient of 0.7394.Compared with other cerebrovascular segmentation methods,the segmentation method has obvious advantages. |
PDF Full Text Request