A Method for Automatic Monoplane Angiography Segmentation

The diagnosis of stenosis, characterized by narrowing the lumen on arteries, requires the inspection of medical images acquired using technics such as X-Ray angiography. Recently, convolutional neural networks (CNN) have been successfully applied to automate segmenting arteries on angiographic images. The main challenge of using these models relies on counting many images where the vessels of interest have been manually annotated. Besides being difficult and expensive to produce, obtaining consent for its use in clinical investigations is even more complicated. Thus, this work presents an automatic angiography segmentation method that does not rely on a CNN architecture. Our results indicate the feasibility of using the proposed method for segmenting vascular components on angiographies with accuracies comparable to other CNN-based state-of-the-art methods.