Magnetic Resonance Image Segmentationof Leg And Knee Joint Based On Live Wire

Knee joints and legs are the most important locomotive organs of the human body,and they are prone to lesions and vulnerable to injury.Magnetic resonance imaging(MRI)is the primary imaging technique currently used to examine knee and leg organs.By segmenting the magnetic resonance images of the knee and leg,the target tissue can be extracted from it,providing valuable information for the physician to diagnose the condition,plan the surgical plan,and estimate the specific absorption rate in the electromagnetic simulation.At present,the semi-automatic segmentation method such as region growing method is used to segment the magnetic resonance images of the knee joint and the leg,which is susceptible to low signal-to-noise ratio.The rapidly developing deep learning method requires a large amount of training data,but the data acquisition is more difficult.These factors will affect the segmentation to a certain extent.In this paper,the Live wire algorithm is used to segment the image.The advantage of this algorithm is that the human-computer interaction segmentation accuracy is high,and the dependence on signal-to-noise ratio is relatively small.This paper mainly optimizes the traditional Live wire algorithm as follows:(1)For the knee joint,the local cost function is improved by algorithm,and the Canny edge detection operator is used to replace the previous Laplace operator,which improves the accuracy of edge localization and reduces influence of the noise on segmentation.Based on the characteristics of Tl and T2 weighted images,the gradient amplitude function is optimized to improve the accuracy of segmentation.(2)Replacing the previous accumulated path cost with the average path cost,so as to reduce the need for the number of seed points at the boundary with larger curvature.(3)In order to improve the segmentation efficiency,in the boundary search,the boundary points determined in the previous round are directly put into the predefined linked list,as a priori condition for finding the next new boundary point,and it is no longer necessary to traverse the entire image.It is also proposed that in a round of calculations,multiple points can be processed simultaneously,reducing the total computing time.(4)For the segmentation of the leg,the Gaussian weighted Euclidean distance and nonlinear interpolation method are adopted,and the information of the key layer is used to automatically predict the tissue contour of the middle layers.This method has certain high segmentation precision and reduces the artificial participation.