| Nowadays,one of the main causes that cause the mortality to have a serious increase in the global population is the Cardiovascular Disease(CVD),and the pathogenesis of Cardiovascular Disease mainly comes from coronary atherosclerosis.Calcified plaques that cause coronary atherosclerosis are important factors that determine and influence the treatment of cardiovascular diseases.At present,doctors use the manual marking to detect the coronary plaques and the process of manual marking is tedious and time-consuming,and there are serious differences in the identification and judgment of coronary plaques by different doctors.It is necessary to propose a method that can automatically segment,recognize and extract coronary atherosclerotic plaques so that the diagnostic efficiency and accuracy of coronary atherosclerotic diseases can be effectively improve.Computed Tomography Angiography(CTA)as a highly reliable and non-invasive diagnostic method for cardiovascular disease can be used to accurately assess CVD.It can be used to accurately evaluate CVD.CTA is considered to play a key role in guiding preventive treatment strategies in the future.However,due to the complex tissue structure of coronary arteries and the uneven gray scale of CTA images,as well as the variable shape and small area of coronary plaques,the automatic detection of coronary plaques still faces great challenges.This paper focuses on the deep learning theory to conduct automatic segmentation and detection of calcified plaque in CTA images of cardiovascular patients.At the same time,the detection effect of calcified plaque can be improved in a complete and accurate way.The following is the main research of the paper:(1)A method based on AG-DenseUNet for the detection of calcified plaque is proposed to solve the problem that the small plaque is difficult to be completely detected.Firstly,this method uses image preprocessing to determine the approximate location of each calcified patch area,and then sends the processed image to the training network of AG-DenseUNet.On the one hand,we adopt the method that keeping each layer of the input and output channels of the neural network is consistent in DenseUNet.The aim is to reduce the loss of image data information and ensure the integrity of the spatial information and feature information in the transfer process.And by improving the DenseBlock module and down-sampling path in DenseUNet,it is more suitable for the detection of cardiovascular lesions.On the other hand,Attention Gate can suppress the interference of irrelevant features during model training,making the model automatically focus on the extraction of calcified plaque features.(2)The differences of two different types of deep learning techniques in the detection of calcified plaque in the whole CTA image of cardiovascular lesions are studied.First of all,the latest research on semantic segmentation networks and attention mechanisms since 2014 is comprehensively introduced,including five types of semantic segmentation networks and seven types of attention mechanisms.In order to comprehensively analyze and compare their advantages and disadvantages,they are used in the identification of calcified plaques in cardiovascular lesions.Firstly,different types of semantic segmentation networks and attention mechanisms are used to detect calcified plaques under the same number of training iterations.then,the same type of network was compared under different training iterations.Then,the influence of using the same network on the segmentation and detection of calcified plaque are compared under different training iterations.Finally,the experimental results prove that Deeplabv3+ has the highest detection accuracy and precision for calcified plaques.Dual Attentional mechanism has the best detection effect on calcified plaques in CTA images,while PSA occupation is the lowest.In addition,experiments have also verified that gender and age have a certain influence on the detection of calcified plaques. |
PDF Full Text Request