Research On Carotid Artery B-ultrasound Plaque Recognition Based On Deep Learning

Cardiovascular and cerebrovascular diseases are a common disease that poses a serious threat to humanity.Their high incidence,disability,and mortality rates are daunting.According to statistics,the number of deaths from cardiovascular and cerebrovascular diseases worldwide each year is as high as 15 million,ranking first among various causes of death.Carotid atherosclerosis(CAS)is closely related to cardiovascular and cerebrovascular diseases.Therefore,accurate identification of carotid plaque in B-ultrasound images is of great significance for timely treatment of patients at risk of stroke.At present,the mainstream diagnostic method in clinical practice is traditional two-dimensional ultrasound.Doctors use probes to obtain images,which has strong subjectivity and poor repeatability,and there are cases of missed detections.Therefore,there are generally errors in two-dimensional ultrasound diagnosis.The different personal operating techniques and vascular shapes of doctors can lead to differences in the cross-sectional images obtained by B-ultrasound.Different measurement techniques can also lead to different measurement results,which poses a certain degree of difficulty for doctors to judge the condition.This article provides auxiliary means for doctors to accurately identify carotid plaques by identifying,segmenting,locating,and three-dimensional reconstruction of carotid plaques,reducing diagnostic errors to a certain extent.The main research content is as follows:(1)Establishment of B-ultrasound sample set.Using the two-dimensional ultrasound images and videos of the subject’s carotid artery provided by the Ultrasound Department of Inner Mongolia Medical University Affiliated Hospital,calibration was carried out under the guidance of a professional doctor,with a sample set of 5000.(2)Research on carotid artery intima-media target detection based on improved YOLOV5 model.On the basis of the original YOLOV5,the adaptive anchor frame is optimized,attention mechanism is added,the head is decoupled,and the backbone network,non maximum suppression algorithm and loss function are improved.The average precision of the improved YOLOV5 l is improved to 94.77%.(3)Research on image segmentation of carotid artery intima media based on improved 2D-VNet model.On the basis of 2D-VNet,the depth supervision mechanism is introduced,the jump connection is improved,the batch standardization layer is added,and the loss function is modified.The segmentation accuracy of the improved model is improved to 87.95%,and the average Hausdorff distance is reduced to 1.52 mm.(4)Research on diagnostic methods for carotid artery plaques.A complete process was designed from inner and middle membrane detection to inner and middle membrane segmentation(YTV).Compared with the non YTV process,the improved 2D-VNet improved segmentation accuracy by 9.91% under the YTV process.Finally,areas with a thickness exceeding 1.5mm were determined as plaque areas based on the segmentation results of the inner and middle membrane.(5)3D reconstruction of carotid artery plaques,presenting visual 3D images for doctors.Generate plaque depth maps based on three-dimensional reconstruction algorithms to achieve three-dimensional reconstruction of plaques,providing a three-dimensional visualization reference for the specific morphology of carotid plaques in blood vessels.(6)Intelligent diagnostic system for carotid artery plaques.It includes main functions such as inner and middle membrane detection,inner and middle membrane segmentation,plaque lesion recognition,and 3D reconstruction.It can also store patient basic information and historical diagnostic results in a database,providing reference for doctors’ clinical diagnosis.