The Research Of 3D Brain Tumor Image Segmentation Based On Improved Convolutional Neural Network And Continuous Max-flow

Brain tumors are a serious threat to human health.Due to their significant differences in size,shape and location,accurate characterization and localization of brain tumor tissue types play a key role in the diagnosis and treatment of brain tumors.MRI-based brain tumor segmentation has become a hot topic because of its good soft tissue contrast and non-invasive features.The manual segmentation of tumor tissue is cumbersome and time-consuming,and is subject to the subjective consciousness of the segmenter.Therefore,how to segment brain tumors efficiently,accurately and fully automatically becomes the focus of research.Aiming at the problems of small dataset,serious class imbalance and low segmentation accuracy of brain tumor image segmentation,this thesis proposes a two-stage segmentation method,which combines the advantages of convolutional networks and traditional methods.Specifically,the algorithm herein includes three steps.The four modalities of the original MRI images are first preprocessed separately.Next,preliminary segmentation is performed using an improved U-Net CNN respectively containing deep monitoring,residual structures,dense connection structures,and dense skip connections.In this step,the attention mechanism is also applied to the field of brain tumor segmentation for the first time,and an improved convolutional neural network based on attention mechanism is proposed for initial segmentation.The preliminary segmentation results then served as the a priori knowledge for a continuous maximum flow algorithm for fine segmentation of target edges.During the initial segmentation network training process,we adopt a multiclass Dice loss function to deal with class imbalance and successfully prevented overfitting using data augmentation.In order to verify the superiority of the segmentation algorithm proposed in this thesis,the experimental data of this thesis uses Bra TS 2017 open dataset,and a large number of experiments are carried out in each stage of segmentation.Qualitative and quantitative analysis of the experimental results.In addition,it also compares the segmentation accuracy and stability with the more advanced segmentation algorithms in the field of brain tumor image segmentation.Experiments reveal that the mean Dice similarity coefficients of the proposed algorithm in whole tumor,tumor core,and enhancing tumor segmentation are 0.9072,0.8578,and 0.7837,respectively.The proposed algorithm presents higher accuracy and better stability in comparison with some of the more advanced segmentation algorithms for brain tumor images.The 3D brain tumor image segmentation algorithm proposed in this thesis can automatically and accurately segment brain tumor regions in MRI images,which can provide guidance for clinical segmentation.