Design And Implementation Of A Renal Tumor Image Segmentation System Based On Recurrent Neural Network And Attention Mechanism

Medical image segmentation is the classification of medical images at the pixel level,so as to get the category of each pixel.tumors,such as kidney tumors and lung tumors,can usually start anywhere in a normal organ.tumors vary in size and shape due to varying degrees of deterioration,and these uncertainties pose a great challenge in determining the disease and localizing the tumor.doctors observe patients' pathological conditions through CT scan pictures.with a large number of scanning slices taken,doctors' workload becomes very heavy and their work efficiency becomes poor.the knowledge of image segmentation direction can be used to classify each pixel of the medical image,and the precise position of tumor and normal organs on the medical image can be obtained,so as to help doctors make more accurate diagnosis.In this paper,an end-to-end multi-information fusion image segmentation method was proposed,which fused the feature information between "significant regions",the correlation information between "potential features" and the sequence information between section sequences,so as to complete the lesion segmentation of renal tumors.the first method in this paper uses attention mechanism to explore feature information at different levels.the region-based approach is used to find associations between significant regions,such as using the similarity between the left and right kidney regions to facilitate lesion segmentation.the feature-based approach mainly focuses on the complex process of deep feature graph in the convolution process.The second method in this paper adopts the convolutional bidirectional GRU method for slice sequences to effectively capture the sequence information between slices.CT scanning is an uninterrupted and continuous process,and sequentiality is the basic requirement for the construction of physiological tissues.Therefore,the convolutional cyclic neural network is used to explore the sequentiality of slice sequences,so as to find out the potential sequence information.Finally,Python and wx Python were used to design the system interface,and a kidney tumor image segmentation system based on deep learning model was written.