The Research Of 3D Brain Tumor Image Segmentation Integrating Improved Fully Convolutional Neural Network And Hybrid Level Set Method

Brain tumor is a common disease in neurosurgery today.Early screening and diagnosis can tremendously improve the cure rate and reduce mortality.In the brain imaging examination method,magnetic resonance imaging(MRI)has become the mostly widely used modality because it offers unique advantages including the absence of radiation damage and the capacity to distinguish among soft tissues.Reviewing large images is very time-consuming and is prone to misdiagnosis.In order to improve the efficiency of diagnosis and reduce false negative rate,this thesis takes threedimensional(3D)MRI brain tumor images as the research object,and uses the deep learning method and the traditional segmentation method to perform full-automatic and high-accuracy segmentation of the whole tumor,tumor core and enhancing tumor.To enhance the accuracy of segmentation,this study proposes an algorithm integrating an improved fully convolutional neural network(FCNN)and the hybrid level set method.The algorithm first performs bias field correction and grey value normalization on T1,T1 C,T2,and FLAIR MRI images for pre-processing.It then uses a cascading mechanism to perform preliminary segmentation of whole tumors,tumor cores,and enhancing tumors using an improved FCNN based on the relationships among the locations of the three types of tumor structures.This simplifies multiclass brain tumor image segmentation problems into three binary classification problems.At the same time,the improved FCNN adopts anisotropic convolutional kernel,dense connections and multiscale feature-merging to further enhance performance.Model training is respectively conducted on the axial,coronal,and sagittal planes,and the segmentation results from the three different orthogonal views are combined.Finally,hybrid level set method is adopted to refine the brain tumor boundaries in the preliminary segmentation results,thereby completing fine segmentation.In order to verify the superiority of the segmentation algorithm proposed in this thesis,the experimental results are analyzed.Using a ratio of 4:1:1,the experiment divided the brain tumor image segmentation benchmark 2017 dataset into a training set,a validation set,and a test set.The results indicate that the proposed algorithm can achieve 3D MRI brain tumor image segmentation of high accuracy and stability.Comparison of the whole-tumor,tumor-core,and enhancing-tumor segmentation results with the golden standards produced Dice similarity coefficients(Dice)of 0.9113,0.8581,and 0.7976,respectively.The 3D brain tumor image segmentation algorithm proposed in this thesis can automatically and accurately segment brain tumor regions in MRI images,which has practical significance for clinical diagnosis of brain tumor disease.