Research On Multimodal MRI Brain Tumor Image Segmentation Method Based On Deep Learning

With the development of medical imaging technology in recent years,computer image analysis methods have become an important means of axis-assisted clinical diagnosis in the medical field.In clinical diagnosis,the recognition of clinical brain tumor lesions usually requires doctors to perform manual segmentation layer by layer.Manual segmentation technology results in a large workload and low accuracy of lesion segmentation.And the results of multiple divisions by the same doctor may not be the same.Due to the various drawbacks of manual segmentation of medical images,a computer is used to assist doctors in analyzing medical images.Magnetic resonance imaging technology can provide doctors with various anatomical imaging and help doctors quickly locate the location of the lesion.Because brain tumors have multiple regions,complex tumor shapes,blurred edges in each region,insufficient image data,and low resolution of MRI images,these problems bring many challenges to lesion segmentation.In order to solve the above-mentioned problems,research on multi-modal MRI brain tumor image segmentation methods has been carried out.(1)Aiming at the problems of blurred brain tumor boundaries and poor segmentation effects in MRI brain tumor image segmentation,a MRI brain tumor segmentation method(cascaded RR-U-Net)based on cascaded neural network with residual loop structure is proposed.The residual loop module is used to replace the original convolution module to prevent network degradation and achieve full feature extraction.And designed a residual recurrent network(RR-U-Net),which can learn the multi-scale features of different receptive fields to avoid the problem of gradient disappearance caused by the deepening of the network layer;the jump connection structure is added to the network to reduce the low-level details Features are added to high-level semantic features to optimize segmentation results;cascading ideas are added to transform multiple segmentation problems into multiple binary segmentation problems,reducing the amount of network parameters and training time,and improving segmentation accuracy.By comparing the segmentation results of other methods and analyzing the experimental results based on various evaluation indicators,the effectiveness and adaptability of the proposed algorithm are verified.(2)Aiming at the problems of low contrast of brain tumor images,large background interference,more redundant information in the image,and time-consuming training,a MRI brain tumor segmentation method based on the attention mechanism of convolutional neural network(RDAtt-U-Net)is proposed.On the basis of the convolution module,it is proposed to use an improved residual error module(IRB)to construct a network,which can effectively prevent problems such as network degradation and disappearance,and improve the performance of the network.As the convolution kernel scale increases the learning parameters,the training is likely to cause problems such as data overfitting.It is proposed to add a hollow convolution module(DCB),so as not to increase the learning parameters and retain the multi-scale features and more image detail information.The attention mechanism(AGs)is introduced in the jump connection of the network encoder and decoder structure to focus the attention coefficient on the local area,suppress the complex feature response of the background,and improve the segmentation accuracy.By comparing the segmentation results of other methods and analyzing the experimental results based on a variety of evaluation indicators,the performance of the algorithm is better than other comparison algorithms,and higher segmentation accuracy can be achieved.