Convolutional networks for segmentation of large vein images

The article presents the results of work on image segmentation individual images of magnetic resonance imaging of the retroperitoneal space. The issues of detection and segmentation of objects the main veins of retroperitoneal space based on the convolutional architecture of a neural network for semantic pixel segmentation are considered. An automatic, accurate and reliable method using the convolutional neural network U-Net for extracting vein vessels from MRI images is proposed. Deep network training with a large receptive field U-Net allows you to achieve significant results even with the presence of low-quality source data, on small training samples. The data expansion strategy seems to be an effective way to reduce the degree of retraining in the recognition of medical images — veins