Meniscus Segmentation And Tearing Grade Diagnosis Based On Deep Learning

The knee meniscus can cushion the impact of the lower limbs and provide joint flexion and extension necessary for movement.Meniscus tear and other injuries will aggravate the instability of the knee joint and make it difficult for the patient to walk,so the early diagnosis of meniscus tear is extremely important in medical research.As an advanced medical imaging method to judge meniscus tear,MRI can present the intact damaged condition of meniscus area and surrounding tissue.However,due to the limited medical conditions and experience in some remote areas,it is impossible to accurately judge the degree of complex meniscus tear,which may lead to misdiagnosis and late diagnosis.With the rapid application of artificial intelligence in computer vision,the methods with deep learning as the main trend gradually show advantages in medical imaging.Therefore,this paper adopts knee MRI data from Tianjin Hospital and Affiliated Hospital of Anhui Medical University to propose a deep learning convolutional neural network based automatic segmentation of knee meniscus and tear grade judgment method.Specific research results are as follows:(1)In order to improve the segmentation accuracy,a Multiscale-Net method for meniscus segmentation of knee joint is proposed.Combined with the convolution layer and pooling layer of the visual geometry network and the decoder part of U-Net network,the 3*3 convolution layer connected by the encoder and decoder is replaced by an improved spatial convolution pooling pyramid module.Finally,the real data set of clinical patients provided by the First Affiliated Hospital of Anhui Medical University was verified and compared with U-Net and U-Net models introduced with ASPP module.The experimental results show that the intersection ratio and Dyess coefficient of the proposed method are 91.25% and 94.89%,respectively,which are higher than other comparison methods.(2)In order to analyze and identify the meniscus tear grade specifically,this paper proposes the improved convolutional neural network based on ResNet-50 residual module and attention mode mechanism to classify and diagnose the four grades of meniscus tear specifically.In this model,lightweight coordinate attention and efficient attention modules are used to improve the attention of the network to the torn meniscus region,and the meniscus region image is automatically segmtioned to extract the region of interest.Experimental results show that the classification accuracy of the improved model is 97.45%,which is excellent compared with the network,and can assist doctors to determine the grade of meniscus tear in clinical practice.