Research On Diagnosis Of Carotid Artery Stenosis And Recognition Of Plaque Calcification Component In CTA Image Based On Different Algorithms

Background and purposeThe epidemiological survey of stroke in China showed that the prevalence rate and morbidity of ischemic stroke in our country are increasing year by year,and carotid artery stenosis is the important cause of ischemic stroke.The accurate diagnosis of carotid artery stenosis is of great significance in the face of the increasing economic and social burden of carotid artery stenosis stroke.At present,CT angiography(CTA)is one of the main non-invasive imaging methods for the diagnosis of carotid artery stenosis.However,the variety of carotid plaque morphology,the subjectivity of diagnostician in judging the degree of stenosis and the increasing amount of CTA examinations for carotid stenosis all bring some limitations to the traditional CTA method in the diagnosis of carotid stenosis.The semi-automatic or automatic diagnosis of carotid artery stenosis in CTA images based on different algorithms has become a research hotspot in the diagnosis of carotid artery stenosis.On the one hand,the computer-aided diagnosis of carotid artery stenosis can solve the problems of large number of patients and large amount of carotid CTA data through rapid identification of carotid artery stenosis lesions.On the other hand,it can identify the characteristics of complex carotid artery plaques quantitatively to avoid deviations caused by subjective factors.Methods1.The imaging data of 45 patients undergoing head and neck CTA and DSA in the same period from June 2019 to September 2019 were analyzed retrospectively in the study of semi-automatic diagnosis of carotid artery stenosis in CTA images based on random walk algorithm.The diagnosis of carotid artery stenosis was selected at the narrowest part of the carotid artery,and DSA diagnosis results were taken as the research criteria.At first,the median filter was used to remove the image noise in the axial CTA image at the same position,and then the boundary of the carotid artery was extracted semi-automatically by the random walk algorithm.Furthermore,in order to obtain the binary images of the carotid artery and the lumen,the threshold method was applied.Finally,the degree of carotid artery stenosis was calculated in comparison with the diagnostic results of DSA.2.The CTA image data of 250 patients,who underwent the examination from January 2019 to January 2020 in the affiliated Hospital of Hebei University,were analyzed retrospectively in the study of automatic recognition of carotid plaques calcification in CTA images based on U-net algorithm.CTA images of 5 patients were excluded due to incomplete or unclear data,and CTA data of 245 patients were eventually included.Two neuroradiologists,who have more than 5 years working experience,selected 490 CTA axial images from the CTA data of 245 patients to form a training data set containing 440 CTA images and an experimental data set containing50 CTA images.The training data set was labeled by the two neuroradiologists for carotid canal wall,lumen and plaque calcification manually.The training data set was used for U-net neural network to deeply learn the characteristics of carotid plaque calcification.Then,the trained U-net neural network was applied to the experimental data set to identify and segment the carotid artery calcification.Finally,the area and maximum thickness of the calcification were measured in comparison with the results of manual labeling calcification.Results1.Semi-automatic diagnosis of carotid artery stenosis in CTA images based on random walk algorithm was highly consistent with DSA diagnostic results,with kappa value of 0.81,and the diagnosis of mild to moderate carotid artery stenosis was more consistent with DSA,with kappa value up to 0.84.Compared with traditional CTA,the sensitivity and specificity of diagnosing mild,moderate and severe carotid artery stenosis were improved,which were 0.78,1.00,0.97,0.81,0.82 and 0.97 respectively.Especially in the diagnosis of moderate to severe carotid stenosis,the random walk algorithm had higher diagnostic efficiency than other semi-automatic methods,and the area under ROC curve was 0.97,P<0.05.2.Automatic segmentation of carotid plaque calcification in CTA images based on U-net algorithm had high efficiency,and taking artificial labeling calcification as the standard,the Dice coefficient was 0.83.The measurement results of carotid artery calcification area and maximum thickness were highly consistent with those of manually labeled calcification,Spearman correlation coefficients were 0.97 and 0.93,respectively,and P were both less than 0.01.Conclusion1.Semi-automatic diagnosis of carotid stenosis in CTA images based on random walk algorithm is a reliable method,which can improve the diagnostic efficiency of traditional CTA.2.Automatic identification and segmentation of carotid plaque calcification in CTA images based on U-net algorithm can be used as an alternative way to mark calcification manually,and it can provide doctors with characteristic information such as calcification area and maximum thickness.3.The algorithm in this paper can not only save manpower and time cost,but also provide reliable basis for formulating treatment strategies and choosing surgical methods and equipment.