Automatic Diagnosis Of Knee Joint Based On MRI

In recent years,the development and application of deep learning in the field of natural images has laid a solid foundation for the development of medical images.A large number of medical models based on deep learning have been proposed and achieved good results.However,due to the particularity of some medical tasks,new challenges are presented to deep learning.Knee joint diagnosis,because of its complex internal structure.The existing deep convolutional neural network structure suitable for natural images does not take into account the specificity of the knee joint in medicine,and it performs poorly when applied to knee joint diagnosis.This article proposes two models for the particularity of the knee joint itself,referring to the doctor’s method of diagnosing the knee joint.Aiming at the complexity of the internal structure of the knee joint,this paper proposes a knee joint diagnosis network based on the positioning of the tissue structure.By defining auxiliary tissue positioning tasks,and training the tissue positioning auxiliary network on the tissue classification data set,the initial positioning of the knee joint tissue is realized.Subsequently,the expert information of tissue location is introduced into the knee joint diagnosis network,and the tissue positioning guide model focuses on the corresponding area of tissue structure,simulating the diagnosis process of the doctor,thereby improving the diagnosis effect of the model.Through the comparison experiment of positioning generation and the comparison experiment of combination method,the validity of the structure proposed in this paper is verified.Comparative experiments with other similar models show that the model proposed in this paper is superior to other models in AUC index.Finally,through visualization,the effect of the tissue positioning auxiliary network on the tissue positioning task is demonstrated,which can relatively accurately identify the meniscus and cruciate ligament.Aiming at the multi-slice data features of the knee joint dataset,this paper proposes a knee joint diagnosis network based on the attention mechanism.This paper first proposes a cross-slice attention structure,which calculates the correlation of image features between multiple slices through the attention structure,and aggregates the image features of multiple slices to achieve collaborative processing of multi-slice data.Through this structure,the diagnosis process of the doctor combining multiple slices is simulated,and the diagnosis accuracy of the model is improved.Subsequently,in response to the problem of sample imbalance,this paper modified the loss function to use weights to adjust the learning rate.By designing ablation experiments,the effectiveness of the attention structure is verified.Through comparative experiments with other similar models,the superiority of this model is verified.Finally,using the method of attention visualization,it is verified that the attention structure can learn the correlation between the features of each slice.