Vessel Identification and Tracing in DSA Image Series for Cerebrovascular Surgical Planning

Brigham And Women'S Hospital, Boston MA

Investigators

Linked publications, trials & patents

Abstract

Project Summary Although Digital Subtraction Angiography (DSA) is the most important imaging for visualizing cerebrovascular anatomy, its interpretation by clinicians remains difficult. This is particularly true when treating arteriovenous malformations (AVMs), where veins and arteries are entangled and need to be carefully identified. The presented project aims at enhancing DSA image series to remove this difficulty. Our long-term goal is to contribute toward the development of intuitive and interpretable visualization tools to improve the diagnosis, planning and treatement of neurovascular pathologies. Our overall objectives in this project are to (i) develop a new method, based on machine learning, to localize the AVM and distinguish between veins and arteries surrounding it in DSA image series, and (ii) develop an algorithm that classifies arteries as terminal, en passage or bystanders. In addition to examine the impact of our approach in planning neurovascular surgeries through a retrospective study. The rationale for this project is that such technology will likely enhance DSA imaging and provide an interpretable tool to clinicians that will facilitate planning cerebral AVM procedures, and furthermore, provide a decision support tool that can be used during surgery to help review and correlate the anatomic findings seen in the surgical field to the preoperative angiogram. To attain the overall objectives, the following two specific aims will be pursued: (1) develop an image processing algorithm for AVM localisation and artery/vein classification and (2) develop an algorithm that can identify arteries as terminal, en passage or bystanders. Under the first aim, we will test our working hypothesis to show that it is possible to localize an AVM in DSA image series and to distinguish between feeding arteries and draining veins surrounding or creating the entanglement, using deep neural networks to outline the shape of the AVM in the images and independent component analysis to understand blood flow disruption. For the second aim, we will establish a set of rules to classify arteries contributing or not to the AVM and implement these rules into a dynamic instance segmentation algorithm that will trace vessels individually, in a DSA image series. This algorithm with rely on a foreground/background subtraction to genrate a vascular graph and on deep neural network to classify vascular junctions to produce an instanciated graph. Using this graph and the pre-defined rules it will be possible to visually distinguish between the different artery patterns. The proposed project is innovative because it will be possible to automatically distinguish veins from arteries and classify arteries as terminal, en passage, or bystander in a DSA image series without altering standard clinical routines. The proposed project is significant because it will enhance DSA imaging with an intuitive visualization allowing clinicians to better understand AVM-induced vessel entanglement in order to preserve vessels from being mistakenly clamped during surgery, thus avoiding intraoperative hemorrhaging or postsurgical deficits. These results are expected to have an important positive impact because they will ultimately provide new opportunities for the development of novel planning techniques to improve the treatment of neurovascular malformations.