Low-dose DSA Algorithm Research Using Convolution Neural Network

The goal of the low-dose digital subtraction angiography(DSA)algorithm is to reduce the radiation dose of patients and to obtain high-quality DSA vascular subtraction images without motion artifacts.The existing research on low-dose DSA algorithm is mainly to remove the artifacts and avoid re-shooting the exposed radiation.The processing results show that the blood vessel details are missing and take a long time.There are great limitations.Convolutional neural networks have powerful nonlinear mapping capabilities,and automatically extract global semantic features and local edge features of images.In recent years,deep learning has developed rapidly in the field of medical image processing.Based on deep learning,this topic aims to reduce the radiation dose of DSA vascular subtraction,and provide powerful assistance for the interventional treatment diagnosis and treatment,while removing the motion artifacts of DSA vascular subtraction and reducing the radiation dose received by patients.The relevant research of this topic is as follows:Based on the dense residual network,the motion artifacts of the DSA image are removed,the quality of the DSA image is improved,the number of clinically taken DSA is reduced,and the radiation dose is reduced.The main idea of the algorithm is: during the training phase,the original background frame(mask)of the artifact-free DSA image is simulated and deformed,and the original contrast frame(live)and the deformed mask are subtracted to obtain the corresponding artifacts.DSA image.The dense residual network mitigates the gradient dispersion and makes full use of different hierarchical features to learn the nonlinear mapping of DSA artifact images to high quality artifact-free DSA images.Simulated DSA image artifacts and real DSA image artifacts were tested separately.Comparative experiments show that the dense residual network model can remove bone tissue artifacts more effectively than the dense connection network and residual network,better maintain image contrast and retain more blood vessel details.The DSA vascular subtraction is generated based on the improved U-net single-frame angiography image,which removes the dependence of the background frame(mask)and reduces the exposure time and radiation dose.The main idea of the algorithm is to use the artifact-free DSA vascular subtraction image and the corresponding angiographic image(live)as the data set.The improved U-net network was designed to learn the end-to-end mapping of single-frame angiography images to high-quality artifact-free DSA images and explored factors that affect model performance.The quantitative and objective evaluation of the network model processing results shows that the cavity residual unit improves the network’s receptive field and extracts the DSA vascular subtraction from the global information of the single-frame angiography image.The loss function using a combination of L2 norm loss and gradient loss can output finer blood vessel details and better maintain contrast.