Research On Image Segmentation Algorithm Based On Multimodal MRI Brain Tumor

Tumor is one of the common diseases which endangers human health in life.The task of brain tumor image segmentation is a hotspot in medical image processing.The purpose is to locate the specific location of the tumor in the brain tissue and accurately extract its contour,it is important and significance for clinical medicine in radiology planning.Because of the complex structure of the brain,such as different shapes and positions,uneven internal gray scale and blurred edges,which make brain tumor segmentation as a challenging task.With the continuous development of medical imaging,the computer image processing technology can be used to locate the tumor accurately in the image,and further analyze the tissue components in the tumor lesion area.The soft tissue anatomy displayed by MRI imaging is more realistic,it is strong discrimination between normal tissue and diseased tissue.This paper proposes two segmentation methods for MRI brain tumor image segmentation,it mainly includes the following two tasks:This paper proposes an automatic brain tumor image segmentation algorithm is that uses region growing with the co-constraint of intensity and spatial texture.Owing to the inhomogeneous structure and blurred boundary of brain tumors,the related image segmentation is not always ideal.The model improves the selection of seed points and the growth rules.Determining the initial seed point candidate set by using multi-mode MRI image fusion,and then utilizing the filtered window,so that the seed point may be automatically selected.Furthermore,to retain the local feature and the boundary information of the tumor,a spatial texture feature is constructed.The proposed method not only outperforms other segmentation algorithms in terms of accuracy but also has lower computational cost,The segmentation accuracy of this algorithm is above 97%.This paper proposes a 3D full convolutional neural network model,which implements three subtasks in the brain tumor segmentation task.The internal structure category problem is not solved in the traditional method.Thus,this paper proposes a multi-modal 3D full convolutional neural network model to segment the brain tumors and their internal structures.Deep learning has the advantage of multi-classification tasks,although there are some problems such as low precision and time consuming in the existing end-to-end deep learning images segmentation methods.In this paper,a multi-modal 3D full convolutional neural network is proposed to solve the above problems,it performs well in three sub-tasks: tumor segmentation,tumor segmentation and tumor edema segmentation.The algorithm firstly normalizes the four modal MRI brain tumor images of FLAIR,T1,T1 C and T2,which uses the histogram to enhance the data,and then inputs the 3D image into the neural network to train.The pyramid pooling module is added to the network model to perform multi-scale information fusion on the features extracted by the neural network.Finally,the deconvolution is performed that restore the image as well as the original size.The output is the result of multiple segmentation of the tumor by analyzing 2 output channels.Experiments was implemented showed that the proposed uses a small number of training samples,And the network has a fast convergence speed,which has been greatly improved in terms of accuracy.The dice of the complete tumor segmentation is above 90%,and the core and edema zone is more than 80%.This paper realizes the segmentation of brain tumor images from traditional methods and deep learning methods,and achieves accurate segmentation results,which has important guiding significance for clinical medicine.