Detection Of Carotid Intima And Media Thicknesses Based On Ultrasound B-mode Images Clustered With Gaussian Mixture Model

The prevalence of cardiovascular and cerebrovascular diseases caused by atherosclerosis is high,and the resulting acute cerebral stroke,coronary heart disease and other atherosclerotic cardiovascular and cerebrovascular diseases have a high clinical disability rate and clinical mortality,which has seriously threatened human life and health.Carotid atherosclerosis is characterized by obvious thickening and hardening of arterial wall,which is the main cause of cardiovascular diseases such as acute coronary heart disease and stroke.Therefore,timely detection of risks and influencing factors related to the severity and causes of carotid atherosclerosis is of great significance for effectively reducing the clinical morbidity of various cardiovascular diseases.The latest ultrasound research results show that the thickness of the intima,media,and adventitia of the vascular wall can be independently characterized and evaluated to assess the degree of atherosclerosis such as coronary heart disease,myocardial infarction,and stroke.However,due to the adverse effects of uncertainty factors such as ultrasound images,severely limiting the carotid artery trilateral membrane separation measurement mode of popularization,and the existing detection method requires a large number of labeled samples for the training of the model,the accuracy of the test results must also rely on the accuracy of the sample calibration,and its detection precision is not high.Therefore,this thesis proposes a method clustering processing based on Gaussian mixture model(GMM)B-mode image to improve the detection accuracy of the carotid artery intima and media thickness.First,based on GMM,the grayscale features of carotid artery modeled,and the probability density function distribution of the carotid artery lumen three membrane and surrounding tissues is obtained.Based on this,a probability model is established,and then the EM algorithm is used to estimate for each category for calculating grayscale average and variance of the parameters,and the characteristic parameter values mapped to image space;Then the gray threshold method was used to detect the boundary between the intima and media of the vessel wall to estimate the thickness and the probability density function distribution of the carotid artery lumen,trilateral membrane and surrounding tissues is obtained.Based on this,a probability model is established.The thickness of the intima and media of the vessel wall estimated by GMM clustering were(0.167 5±0.016 4)mm and(0.227 8±0.020 8)mm,respectively.Compared with the results of direct gray threshold estimation(0.174 2±0.020 8)mm and(0.221 1±0.02 2)mm,the standard deviation decreased by 21% and 5%,respectively.Compared with the results of manual fine measurement by two professional doctors,the average error of intima and media thickness measured by the proposed method were reduced by 7.5% and 8.7% respectively.In summary,the method of this thesis has a higher accuracy in measuring the intima and media thickness of the carotid blood wall,which is helpful for clinicians to timely understand the symptoms and physiological conditions of cardiovascular diseases caused by related diseases.In addition,the detection process does not require large sample labeling,and the applicability is strong.The consistency of the measurement results of different given thresholds is also good,which is conducive to reducing the measurement errors introduced by clinical testers and technicians due to subjective factors,and improving the diagnosis and clinical treatment level of clinical testers and doctors for cardiovascular and cerebrovascular diseases.