Research On Automatic Detection Method Of Vulnerable Plaque In Intravascular Optical Coherence Tomography Images

Coronary artery disease(CAD)is the most common cause of death in the world.Acute coronary syndrome(ACS)is the most dangerous condition for CAD.Studies have demonstrated that nearly 70% of ACS events are caused by the rupture of coronary atherosclerosis vulnerable plaque.Therefore,it is important to detect these vulnerable plaques for the diagnosis and treatment of CAD.At present,the highest resolution imaging technique for CAD is intravascular optical coherence tomography(IVOCT).One IVOCT imaging may produce a large number of images.The traditional method of vulnerable plaque detection for these images is that cardiologists diagnose whether there are vulnerable plaques on IVOCT images according to their experience,such a manual method is time-consuming,laborious,and subjective.Therefore,it is necessary to develop an automatic detection method of vulnerable plaques for IVOCT images.Currently,the deep learning method has been applied to the automatic detection of vulnerable plaques in IVOCT images.However,these studies still have defects such as inappropriate data pre-and post-processing methods and unstable performance of single deep neural network.In order to solve the problems above,this dissertation presents an automatic detection method of vulnerable plaque for IVOCT images,which consists of IVOCT image pre-processing algorithm,post-processing algorithm,and neural network integration method.It provides technical support for the automatic detection of vulnerable plaque for IVOCT images.The dissertation first introduces the characteristics of experimental data and the evaluation metrics for vulnerable plaque detection.Then,the pre-processing algorithms for IVOCT image are studied,and the algorithms for the detection of intracavitary contour of the vessel wall and data augmentation are proposed.Next,a post-processing algorithm for vulnerable plaque detection is designed.Finally,a neural network ensemble method is proposed to improve detection performance through an ensemble of multiple neural networks.This dissertation also verified the proposed algorithms and evaluated the performance of vulnerable plaque detection according to the evaluation metrics.The main research results of this dissertation are summarized as follows.(1)In order to solve the problem of redundant information of device and blood in IVOCT images and the problem of small sample size,an IVOCT image pre-processing algorithm is proposed.The algorithm utilizes the characteristics of the polar coordinate system and the Cartesian coordinate system to remove redundant information of devices such as the calibration circle,imaging catheter,protective sheath,and guide wire;it also removes blood information in the vascular lumen through the edge detection algorithm for vessel wall,thereby eliminating their influence on the detection of vulnerable plaque.In addition,the algorithm increases the number of samples variously by means of data augmentation methods such as end-to-end connection,resampling,and flipping,which helps the neural network to learn the features of vulnerable plaques.(2)Based on the characteristics of IVOCT images and the proposed pre-processing algorithm,a post-processing algorithm for vulnerable plaque detection is proposed.The algorithm performs post-processing on the vulnerable plaques output by the neural network by means of union of intersecting regions,duplicated region processing,and small gaps removal to improve the performance of vulnerable plaque detection.(3)In order to solve the problem of unstable performance of single deep neural network,a vulnerable plaque detection method based on neural network ensemble is proposed.First,the method selects different networks of object detection to detect vulnerable plaques of IVOCT images respectively;then,a two-stage ensemble operation is conducted by the method: the classification ensemble is applied to images in the first stage and then the regional ensemble is applied in the second stage according to the classification ensemble results.The result of the ensemble is output as the final vulnerable plaque detection result.The method proposed in this dissertation was evaluated on the test set.The results show that our method can achieve an accuracy rate of 88.84%,a recall rate of 95.02%,an overlap rate of 85.09%,and a detection quality of 88.46%.Experimental results show that the proposed method can be used as an auxiliary diagnostic tool for vulnerable plaque detection in IVOCT images,thus promoting the clinical application of IVOCT images in the diagnosis and treatment of CAD.