Automated Measurement Method Of Common Carotid Artery Intima-media Thickness In Ultrasound Image Based On Deep Learning

The common carotid artery(CCA)intima media thickness(IMT)is one of widely accepted and important markers of early atherosclerosis.In clinic,ultrasound imaging technique is widely used to obtain carotid artery images.However,IMT based on manual tracing is time-consuming,complicated and subjective.Therefore,we propose and study a new IMT measurement algorithm based on deep learning which is automatic,fast,accurate and robust.The algorithm is mainly divided into three steps.Firstly,Convolutional Neural Network(CNN)is applied to identify carotid artery distal and extract region of interest(ROI),which includes intima-media complex(IMC).Then,the different pattern classifiers based on the stacked auto encoder,SAE_NB and SAE_LM,are added to classify pixels of ROI.Pixels in ROI are divided into non-boundary pixels and boundary pixels by SAE_NB.Then,by SAE_LM,the latter will be classified as Lumen-Intima Interface(LII)pixels and the Media-Adventitia Interface(MAI)pixels Finally,the classification results are chosen according to the location information of the classified regions,and choose the reliable classification regions based on region area and region center.Extract final boundary with the method of curve fitting based on mean square error minimization and complete the task of IMT measurement.The image database provided by Sepp Los Institute of neurological medicine is for IMT measurement of carotid ultrasound images.The Gold Truth(GT)of the IMT was manually measured a total of four times by two experts and then averaged.The automatic segmented(AS)IMT was computed using the proposed method.Experiments show that,the mean of the absolute error standard deviation between AS and GT IMT was 13.3 20.5 m,and the correlation coefficient was 0.9907 between the two.The validity of the proposed algorithm in this thesis is verified.