Research On Medical Image Segmentation Method Based On Convolutional Neural Network

In recent years,in the field of image processing,medical image segmentation algorithms have become one of the research hotspots of scholars.It can extract the region of interest in the target image,which will help alleviate the current medical situation in my country and provide medical workers with help.Brain tumor is one of several malignant tumors with high morbidity and mortality in my country,and it is difficult to cure.Surgery is currently the main treatment method.Magnetic resonance imaging(MRI)is an important means to assist in the diagnosis and treatment of brain tumors.Precise segmentation of brain tumor lesions and tissues plays an important role in assisting diagnosis and treatment.Because brain tumor MR images have the characteristics of low contrast,many noise points,blurred boundaries,irregularities,etc.,and they are easily affected by factors such as uneven magnetic field and local human motion in the imaging,and the pixel resolution of MRI equipment is limited,it will cause The boundary pixels of different tissues are stacked together,causing the edge volume effect,so that the traditional manual reading method is time-consuming and labor-intensive.Therefore,the automatic intelligent segmentation algorithm with the help of various image segmentation algorithms is particularly important.For this purpose,this article combs the segmentation methods and methods of brain tumor MR images in recent years,and conducts in-depth exploration and understanding of their basic ideas,methodologies,advantages and disadvantages.A multi-scale feature fusion full convolutional neural network brain tumor MR image segmentation model is developed.This algorithm introduces multi-scale input and feature fusion,strengthens the network's low-level and high-level feature extraction capabilities,and uses a jump structure,The fused features are used as part of the input and final result of the first path of the network,and convolution kernels of different sizes are used in different paths of the network to achieve the purpose of multi-scale feature input and extraction.This paper uses the training set of the Bra Ts Challenge to verify and test the network model,and perform normalization and other preprocessing on the four modalities of the brain tumor MR image of the dataset.The experimental results show that the MFF-FCN model has good performance in feature extraction and segmentation accuracy,and it takes a short time and is more practical,indicating that these improvements can help improve the network segmentation performance.