Application Of Improved U-net Network In Coronary Artery Segmentation

Coronary artery disease is the number one killer threatening the safety of human life.It is very important for the early prevention and diagnosis of coronary artery disease.The presence of plaque and coronary artery stenosis are the main causes of coronary heart disease.The detection of plaque and the segmentation of coronary artery become the first choice of detection of coronary artery disease.At present,the mainstream method for the diagnosis of coronary artery disease is coronary CT angiography(CTA),which can generate a large amount of coronary image data.By processing the CTA images,the positioning and stenosis grading of coronary arteries can be achieved for the diagnosis of coronary artery disease.CT image processing has become a hot topic in the field of computer aided diagnosis.Image segmentation is an important tool in medical image processing.It can segment images into different regions according to different tissue types and organs for visualization and diagnosis.Nowadays,most medical images come from modern imaging technologies such as CT and MRI,which will generate a large amount of data.If manual or semi-automatic segmentation is performed,the process is time-consuming,tedious and dependent on the experience of clinical experts.In addition,due to the problems of uneven gray scale and fuzzy boundary of CT images,the traditional segmentation method can segment the contour boundary of coronary arteries,but its segmentation accuracy is not ideal.Therefore,it is necessary to study more efficient segmentation algorithms.In recent years,with the continuous development of deep learning,convolutional neural networks have been applied in the medical field.The multi-layer structure of convolutional neural network can extract high-quality features,among which the shallow convolutional layer learns the features of local areas and the deeper convolutional layer learns some abstract features,whichare less sensitive to the size,position and direction of objects,thus contributing to the improvement of segmentation performance.In this paper,three-dimensional convolutional neural network structure is applied to the coronary artery segmentation task.In this paper,the three-dimensional U-net convolutional neural network is used to segment coronary arteries,and its practicability is tested by using multiple data sets in the background of centerless and centerless lines.Firstly,the volume of the coronary CT angiography image input into the network was adjusted to fit the network.When training the network,the number of channels,the number of image layers and the number of samples for each training were adjusted to reach the optimal value.In order to reduce the overfitting problem,the experiment in this paper also selects and tests various data transformations,and USES the normalization method corresponding to its individual for different types of data sets,which is slightly better than the traditional normalization method.The results show that the Dice coefficient based three-dimensional U-net network in this paper is superior to other methods.