Research On Carotid Artery Ultrasound Image Blood Vessel Target Tracking Based On Siamese Network

In recent years,the technology of artificial intelligence in the field of computer vision has developed rapidly.In the field of medical diagnosis,various new technologies of computer-aided diagnosis are also emerging.With the development of technology in the era,medical diagnosis and treatment technology is improving day by day.From the doctor’s traditional diagnosis model,it has gradually developed into a mode where medical imaging equipment assists doctors in diagnosis.Nowadays,various medical imaging technologies have become extremely valuable means in clinical diagnosis.Medical image analysis can improve the efficiency and accuracy of doctors in diagnosing diseases,such as using image segmentation and recognition technology to obtain and recognize the contours of the target location,obtain corresponding clinical parameters,and provide doctors with reliable reference and help.Deep learning technology is good at this.The analysis results of medical images can provide doctors for medical diagnosis,and can assist doctors in quickly locating the lesion area and analyzing the disease.Based on the existing domestic and foreign target tracking methods and medical ultrasound image auxiliary processing system,this article mainly proposes a target tracking method based on carotid artery ultrasound image data set.This paper proposes a Siam FC-Ultrasound Images network for carotid artery target tracking by carotid ultrasound.Its main body structure consists of two parts: the twin sub-network and the classification regression sub-network.The twin sub-network uses a modified50-layer deep residual network(Resnet50)as the backbone network,and the classification and regression sub-network consists of classification branch and regression The branch composition improves the prediction efficiency of network target tracking by reducing the number of layers and reducing the dimension of the feature map without affecting the accuracy.And in order to optimize the shortcomings of low definition of image information and blurry image of ultrasound images,this paper uses an image preprocessing method of gray-scale histogram equalization to improve the target tracking effect of ultrasound images.At the same time,the network is based on the actual project requirements of the engineering system,and uses the method of model tailoring to trim the filter of the model network,which reduces the parameters and convolution in the network without significantly affecting the tracking accuracy of the model.The number of calculations improves the tracking speed of the network,and solves the problems of high storage overhead and untimely system response due to the high calculation times of the feature extraction backbone network model parameters in the twin network.