Research And Implementation Of Brain Tumor Segmentation Algorithm Based On Deep Learning

Tumors seriously endanger the health of residents.At this stage,tumors have become one of the main causes of death among residents in China.With the development of modern imaging technology,especially magnetic resonance imaging technology has greatly helped the diagnosis and treatment of modern medical treatment The segmentation of brain tumors is an important basis for doctors to make a diagnosis However due to the shape,location,structure of brain tumor is complex and there are very large differences in different patients,which makes it difficult to manually segment brain tumors.Manual segmentation of tumors can be very time consuming and can result in different segmentation effects due to the doctor's personal experience Therefore,the research on automatic segmentation of brain tumors has important application valueIn recent years,domestic and foreign research institutions have made many breakthroughs in the field of artificial intelligence,especially deep learning has been widely used in academia and industry.Taking medical image segmentation as an example,the image segmentation method based on deep learning no longer needs to manually design features,and the model itself learns the deep information in the data,which can speed up the diagnosis and accuracy,and avoid misdiagnosisThe main contents of this paper are divided into the following pointsFirstly,this paper proposes a U-Net segmentation model based on pyramid fusion U-Net is a classic model in the field of medical image segmentation.It has two parts,Encoder and Decoder.However,for the details of the tumor and the boundary part,U-Net is difficult to completely segment.Therefore,in order to effectively capture the details of tumors,this paper proposes a U-Net segmentation model based on pyramid fusion to achieve the fusion of multiple scale features,which solves the problem that U-Net can't be segmented for smaller scale tumors.Improving the accuracy of tumor segmentation.Secondly,this paper adopts an Attention-based approach to achieve brain tumor segmentation.In the process of convolution,the extracted features are partially redundant,so allows the model to automatically assign different weights to different features,making the model more concerned with features that contain more tumor information,to provide more accurate information for the final segmentation.At the same time,the attention method also increases the interpretability of the neural network.Finally,the features in the middle of the network are visualized,and the model is given a greater weight to the tumor area in the MRI.Finally,this paper uses Flask to implement an online auxiliary analysis and visualization system based on B/S architecture.The system functions mainly include doctor registration,querying the current number of cases,and whether the cases have been treated.According to the business rules,for patients who have not been treated,the doctor can upload data to the system,and the system automatically completes segmentation.The tumor information is visualized to the doctor,and tumor information in each MRI modality.The system realizes online auxiliary analysis and visualization,which has important application value for clinical diagnosis and treatment.