Three-Dimensional Reconstruction Of Knee Cartilage MRI Image Based On Improved Ray Casting Algorithm

In the process of 3D reconstruction of knee cartilage MRI images,how to segment a given medical image is the basis for 3D reconstruction.At the same time,how to smooth the reconstructed cartilage model without affecting the surface area and volume of the reconstructed model is also a problem that needs to be solved at present.At the segmentation stage,U-Net network has shown sufficient advantages,but the gray value of cartilage is similar to that of surrounding muscle tissue and joint fluid,which is prone to mis-segmentation.Moreover,the knee joint cartilage is narrow,long and small,which is prone to fracture during segmentation,and the soft bone segmentation of the defective part may be ignored.These problems will lead to errors in the reconstruction results and have a wrong impact on the doctor’s diagnosis.In the reconstruction stage,because MRI images have different scanning parameters,the reconstruction results will be relatively rough,while smoothing the cartilage model will lead to excessive shrinkage of the model,which will affect the quantitative analysis of cartilage.For the above three problems,this thesis proposes a new network,and adds a distance-based smoothing method on the basis of the ray casting algorithm.The specific work of this thesis is as follows:Firstly,to address the problem of mistakenly dividing cartilage and other tissues during segmentation,a multi-scale information fusion network model based on U-Net network is proposed.The pyramid module is added to the network.The pyramid module samples the given input with multiple branches and different sampling rates of cavity convolution,which is equivalent to capturing the context information of the given image at multiple scales,increasing the receptive field.Secondly,in view of the insufficient ability of the U-Net network to extract slender knee joints,the U-Net network is enhanced in this article by incorporating a blended attention mechanism,comprising both location attention and channel attention.The segmentation ability of the network for small targets and cartilage defect regions is enhanced by adding attention to the cartilage part of different channels.Thirdly,to address the issue of the coarse surface of the regenerated model,the present study explored a three-dimensional reconstruction technique that leverages the ray casting algorithm in volume rendering to reconstruct the cartilage of the knee joint,and proposed a distance-based smoothing method to prevent excessive shrinkage of the model,which truly restored the knee joint cartilage model,laying a foundation for quantitative research of knee joint cartilage.Finally,so as to verify the advantages of the algorithm in this thesis,this thesis conducts ablation experiments and comparative experiments on clinical medical datasets to verify that the new modules in the segmentation and reconstruction stages in this thesis can improve the accuracy of segmentation and 3D reconstruction to a certain extent.And through animal experiments,the estimation of volume and surface area verifies the validity of the algorithm in this thesis,and proves the accuracy of the algorithm in this thesis,and the error is within the clinically acceptable range.