Research On Coronary Plaque Detection Method Based On Convolutional Neural Network

At present,the incidence of cardiovascular disease in China is accelerating,and the main cause of cardiovascular disease is the presence of atherosclerotic plaques in the coronary arteries.Therefore,accurate diagnosis of coronary plaques is a major health issue that needs to be solved urgently.Coronary CT Angiography(CCTA),proposed in recent years,has gradually evolved into a standard method for early screening of coronary artery disease because of its relatively high resolution and non-invasive,three-dimensional imaging..Traditionally,radiologists’ visual inspection of CCTA images is used to determine whether there are plaques in the coronary arteries.However,each patient may have large individual differences in CCTA images,visual examination methods rely heavily on the clinician’s domain knowledge and are time-consuming and labor-intensive.The current automated method mainly extracts a large number of quantitative features from images to describe the texture and spatial structure of coronary plaques.However,the quantitative features used in this method are mainly carefully hand-designed by computer vision experts.Rely heavily on accurate segmentation of coronary plaque.In response to the above problems,this paper studies two methods based on deep learning to detect coronary plaques,and designs related experiments.The main work of the research is mainly divided into the following two points:(1)Coronary plaque detection method are based on fully supervised learning.This paper studies the precise detection of coronary artery plaque based on the framework of convolutional neural network and recurrent convolutional neural network.First,based on the coronary artery centerline that has been obtained,this paper designs a plan to extract samples along the trend direction of the coronary artery centerline.The extracted sample is a three-dimensional coronary artery block;next,convolutional neural networks and loops are used.Feature extraction of input coronary artery samples through convolutional neural network,and then uses the discriminator to predict the final result;because there are too few labeled samples in the data set,this paper proposes three different data enhancement strategies to solve the positive and negative coronary artery data set.Through comparative experiments between the sample generation scheme and the data enhancement scheme,the final plaque detection accuracy of the two plaque detection frameworks are 81.63% and69.05%,respectively.The experimental results fully prove that the method proposed in this paper is effective in coronary artery plaque.The recognition has a relatively high accuracy.Compared with the traditional method,the method proposed in this paper does not require design of manual features carefully,and the detection result does not depend on the precise segmentation of the coronary artery,which has a relatively high clinical application value.(2)Coronary artery plaque detection method based on semi-supervised learning.The training of the deep learning model requires a large number of labeled training samples for training,while there are only a small number of labeled samples in the coronary artery data,so it is difficult to train a robust model.In response to the above problems,this paper studies a semi-supervised coronary plaque detection method based on consistency training.First,extract the coronary artery samples as the network input according to the sample generation scheme proposed in the previous chapter,and pre-train the two coronary plaque detection models proposed in the previous chapter according to the labeled samples,and save the model parameters with the best verification performance;Then use the saved optimal network model and the consistency training method to generate unlabeled sample pseudolabels,and then mix the pseudo-label samples and a small amount of labeled samples into the patch detection network for re-training,and iteratively realize the task of detecting vein plaques.Compared with the fully-supervised method proposed in the previous chapter,the semi-supervised strategy proposed by this network has improved detection accuracy by 3.86%and 3.46% respectively.The method proposed in this paper provides an automated solution for the detection of small samples of coronary plaques.The large amount of unlabeled coronary data on medical data is used to improve the detection accuracy of the network.It is hoped that the deep learning method can be used on small sample data.application.