Research On Coronary Artery Segmentation Based On Deep Neural Network

Cardiovascular diseases are very common and have become one of the most fatal diseases in humans.Coronary heart disease is a heart disease caused by coronary artery disease and is one of the main cardiovascular diseases.Coronary Computed Tomography Angiography(CCTA)is mainly used in the early screening of clinical diagnosis of coronary heart disease.With the continuous development of computer aided diagnosis technology,diagnostic technology has put forward higher requirements for automatic processing and accuracy of CCTA images.In this paper,CCTA image data is taken as the research object,starting from the image preprocessing step,a noise image super-resolution method based on fractal feature analysis is proposed to obtain noise-free CCTA images with rich texture details.Further,for the problem of automatic high-precision segmentation of medical images,based on target detection and segmentation deep neural network,the coronary artery segmentation based on multi-feature fusion MASK RCNN and the coronary artery calcification lesion segmentation method based on jump link attention Unet are respectively proposed.The heart disease auxiliary diagnosis system provides technical support.The main work of this paper is as follows:Firstly,noise will inevitably be introduced during the acquisition and processing of CCTA images,and noise will affect the feature extraction of the segmentation network.To solve this problem,a noise image super-resolution reconstruction method based on fractal feature analysis is proposed.Fractal analysis can effectively describe the texture characteristics of the image,and the fractal analysis is applied to the super-resolution reconstruction model.Further using the local feature analysis of the image,a denoising method based on the local shape dimension is proposed to restore the noise-free image.On the basis of this research,,multifractal spectrum is used to describe and analyze the local features of the image,and a fractal filtering method based on multifractal spectrum is proposed.Based on the observation of the degradation model,this paper models the interpolation and denoising problems in the same framework to restore the noise-free image.Secondly,the segmentation accuracy of the segmentation network is improved by multi-feature fusion,and a coronary artery segmentation method based on multi-feature fusion is proposed.Aiming at the weak boundary problem of CCTA images,a boundary feature extraction method is introduced to extract effective boundary features.The CCTA image has a complicated background.The shape and gray intensity of some tissues and organs are similar to coronary arteries,such as pulmonary artery,pulmonary vein,and bone,which makes high-precision segmentation difficult.To solve this problem,the fractal dimension can be used to describe the complexity of the image and the nature of the similarity between regions,which can be used to distinguish the coronary arteries from other tissues and organs.Finally,the use of feature fusion methods can complement the advantages of features and improve the ability of the network to perceive detailed information and extract features.Thirdly,using the skip connection structure to integrate the dense attention mechanism into the segmentation network,a Unet network based on dense skip connection attention is proposed.Skip connection can merge the semantic and coarse-grained features of the decoder network with the bottom-level and fine-grained features of the decoder.In this paper,the self-attention mechanism is merged into the Unet network by means of dense jump connections to generate high-precision segmentation results of calcified lesions under the complex background of CCTA images.