Design And Implementation Of Bone Image Analysis System Based On Fine-grained Image Recognition

Bone is one of the most important organs in the human body.With the current aging situation,bone diseases such as fractures and osteoporosis have become the root cause of long-term pain in many middle-aged and elderly people.Traditional skeletal imaging diagnosis is evaluated by experts based on their own experience and standards.It has limitations such as strong subjectivity,low diagnostic efficiency,and certain randomness.The emergence of computer-aided diagnosis system can provide doctors with a rational and efficient diagnosis advice,and effectively reduce the doctor's work pressure.With the rapid development of deep convolutional neural networks in the field of computer vision,deep learning can automatically learn the features and distributions through massive data,so it is more suitable for data modeling and auxiliary diagnosis.However,unlike conventional image analysis,the differences in images of different bone diseases in bone images are small,the analysis is difficult,and more detailed features are needed for identification.Therefore,based on the characteristics of skeletal imaging in clinical diagnosis,this paper from the perspective of fine-grained image analysis,based on musculoskeletal data set,bone age data set,bone density data set,combined with bilinear feature combination and self-attention mechanism information allocation a non-local convolutional neural network model based on self-attention mechanism is designed.The model uses non-local modules to allocate attention in the spatial domain and the channel domain,allowing the model to learn local features in addition to local features in the convolutional layer,which enriches the content of the feature map.Then learn more detailed features.In addition,for the characteristics of the skeletal image with shallow features and fixed structure,Unet is used for image segmentation preprocessing to remove redundant information such as characters and watermarks in the skeletal image,further improve the classification accuracy,and mine difficult samples during training.Perform data enhancement to improve the generalization performance of the model to a certain extent.Cross-validation experiments show that the proposed non-local convolutional neural network model musculoskeletal data set and bone density data set reach F1 values of 72.1%and 88.0%;the mean square error of 1.85 on the bone age data set,better than bilinear CNN model and baseline model.Finally,this paper designs and implements a skeletal image-assisted diagnostic analysis system based on the above model,deploys the model in a Web system for use by users,and conducts performance tests.