CT Image Brain Tissue Segmentation Based On Deep Learning

Segmentation of CT brain tissue plays an important role in the diagnosis such as cerebral hemorrhage,brain trauma and cerebral infarction.It can obtain pathological information,analyze brain tissue size,and help doctors formulate treatment plans for brain diseases.The research of automatically segmenting CT brain tissue base on deep learning techniques,which is extremely important for the diagnosis of brain diseases.Due to the brain tissue region in the clinical CT data is not at the center of the image,the overall idea of positioning first and then segmentation is proposed.First detect the location of brain tissue in the CT image.The coarse segmentation results of brain tissue are obtained by deep learning technology.The centroid of the coarse segmentation is used as the center of the region of interest,in order to obtain accurate positioning information.Then design a deep learning network framework that is more suitable for segmenting CT brain tissue,for finely segmenting the region of interest.It is proposed to add BN(Batch Normalization)layers to the V-Net network structure to improve the robustness of the model.Then,combine with the attention mechanism.Through dynamically adjusting the importance of the channel during training process,the recognition ability of the network to the brain tissue is improved.The new loss function is proposed to avoid the segmentation leakage and segmentation shortage phenomenon,so as to improve the segmentation accuracy of the model.Finally,the tag data is mapped to the original size of CT image through post-processing.The experimental data shows that the proposed deep learning method reaches 0.988 on the commonly used evaluation standard Dice coefficient(the set similarity measure function named by Lee Raymond Dice).Compared with other methods for segmentation of CT brain tissue,it has a greater improvement and stronger robustness.It can accurately segment the brain tissue in CT images,which is helpful for doctors to obtain pathological information and assist in the diagnosis of brain diseases.