Research On Lesion Segmentation And Restoration Method Based On Mr Brain Tumor Image

MRI is a necessary technique in routine craniocerebral examination.Doctors can screen and diagnose disease experiencedly by analyzing MR brain images from patients.High-efficiency and high-accuracy assisted diagnosis technology brings great convenience in disease diagnosis,which can not only greatly improve the diagnosis efficiency of doctors,but also have great significance for the cure rate and survival rate of patients with brain tumors.There is a difference in gray distribution between the tumor area and the normal area in the medical brain tumor image,which is manifested by gray distortion,missing or distorted texture details,etc.,which seriously destroys the consistency of the overall texture structure of the image.It causes challenges in various image analysis areas,such as image registration,atlas construction,and atlas-based segmentation.This article studies from brain tumor image lesion segmentation and image recovery.Combining artificial intelligence technology,two algorithms that can simultaneously achieve lesion segmentation and inpainting of MR brain tumor images are proposed,namely the TLS~2D algorithm and the VAE-GAN network.Both are implemented under non-strong supervised learning mechanism,without the need for a lot of fine manual labeling information,and without relying on seed points or boundary initialization,to achieve automatic tumor segmentation and image lesion repair.And the algorithms can perform images in batch size,which improved the overall efficiency of the algorithm.First,a transformed low-rank structured sparse decomposition model(TLS~2D)is proposed,which is a scheme that integrates structured sparsity constraints,image alignment,and adaptive spatial constraints.Based on the idea of matrix decomposition,the pathological images are decomposed into two parts:the low-rank component as the recovered image and the structured sparse component as pathological segmentation.The algorithm is improved based on the low-rank sparse decomposition model,and the well recovered images can be obtained using TLS~2D with the combined structured sparse and computed image saliency as the adaptive sparsity constraint.Then,based on the basic principle of the deep generative model,combining the structure of the variational autoencoder and the generation adversarial network,an improved semi-supervised lesion segmentation and image recovery algorithm is proposed.The algorithm model is based on the VAE structure and incorporates the discriminator structure in the generative adversarial network.The normal brain image data distribution is introduced during the discriminator training process to improve the quality of the generated image of the generator.Pathological regions can be segmented by evaluating the residual image.The efficacy of these proposed methods is verified on synthetic and real MR brain tumor images.The classic performance evaluation indicators of medical image segmentation and image inpainting are used to objectively evaluate the algorithm results.Experimental results demonstrate that the effectiveness of the proposed algorithm.Compared with other methods,our TLS~2D can effectively provide satisfactory performance and is robust for distinguishing pathological regions.