Intelligent Screening And Lesion Analysis Of Carotid Ultrasound Imaging Based On Deep Learning

Carotid atherosclerosis is one of the important causes of various cardiovascular diseases,and carotid atherosclerosis has been widespread in middle-aged and elderly people in China,so it is of great significance to reduce the incidence of cardiovascular diseases and prolong life by timely screening to detect plaque and intervene and treat symptoms.The use of ultrasound to scan carotid plaque requires subjective identification by a professional doctor,and the doctor’s fatigue and lack of experience will cause errors in plaque identification,and a large number of ultrasound images also lead to overload of doctors,based on the above problems,carotid artery screening and diagnosis can not be effectively popularized.Therefore,it is of great significance to use deep learning to process carotid ultrasound images and establish an intelligent diagnosis system close to the screening process,which can assist doctors in making judgments and improving diagnostic efficiency.However,in actual clinical screening diagnosis,due to the accuracy of ultrasound machines,doctors’ scanning techniques and experience differences,clinicians need to spend a lot of time to screen clearer carotid lumen images for plaque screening.In the previous analysis,there was a lack of lesion research content based on the image of the quality inspection section,and the research method was easily affected by noise confusion information,which increased the false detection rate of plaque.In addition,the current analysis of carotid plaque is mostly limited to a single type of section,which does not provide doctors with comprehensive lesion information.In view of the above problems,this paper studies the intelligent screening and diagnosis technology of carotid artery ultrasound imaging based on deep learning from the perspective of paying more attention to the problems faced by clinicians and closer to the clinical screening process,and develops the carotid artery intelligent auxiliary screening and diagnosis system on this basis.The main research work of this thesis includes:(1)Classification control of carotid artery ultrasound section images.Based on the problems faced by clinicians and the characteristics of ultrasound images,an innovative view of section image classification control before plaque segmentation detection is proposed,three types of carotid artery section image datasets are created,and six classification networks are used for transfer learning comparison,and the DenseNet network using transfer learning is obtained as the optimal classification control model for carotid ultrasound through experimental comprehensive analysis.(2)Classification and detection of luminal plaque on images of longitudinal sections of carotid arteries.Based on the detection of carotid plaque and the determination of the specific location of carotid plaque from the screened longitudinal image in the clinical screening process,a plaque detection algorithm based on the lumenal lumen of the carotid longitudinal section of the carotid artery based on the improved YOLOv7 network is designed,and the network learning ability is enhanced by introducing the CA module and SIoU loss function,and the effectiveness of the improved network proposed in this paper is verified by horizontal comparison experiments and ablation experiments of its own network.(3)Segmentation of luminal plaque on a cross-sectional image of the carotid artery.Based on the analysis of the specific morphology of plaque in cross-sectional images in the clinical screening process,a carotid artery lumen and plaque segmentation dataset is created,and a segmentation and comparison experiment is carried out through ten segmentation networks of classical segmentation network and improved U-Net segmentation network,and the U-Net network is comprehensively analyzed as a model of lumenal plaque segmentation network for carotid cross-sectional images.(4)Design and implement a carotid artery intelligent auxiliary screening and diagnosis system that can be used for clinical screening.Realize the reading,processing and display of patient information,and apply the three types of models obtained from the above experiments to this system through modular design to realize the section classification control of carotid ultrasound images and the classification detection and segmentation of luminal plaque.