The development of new devices and embolic agents has expanded the endovascular approach for cerebrovascular malformation treatment (Pierot et al. 2013), whether in preparation for microsurgery (Natarajan et al. 2008), radiosurgery (Blackburn et al. 2011), or occasionally standalone. The interventionalist must know several characteristics of the malformation in order for the highest chance of successful embolization: the exact vascular supply of the malformation, the number of feeders and feeding pedicles, the flow dynamics of the feeders, and the characteristics of draining vessels or associated aneurysms. Obtaining a clear and exact picture of these characteristics is difficult with especially dense, complex, and very distal lesions, rendering planning and subsequent embolization much more technically challenging.
Any number of feeder arteries may supply a given lesion, comprising two main patterns of flow: the supplying artery either projects a variable number of feeder arteries to the lesion (while the main artery supplies brain tissue past the branch point) or directly supplies the lesion. The former vascular pattern, known as vessel in passage feeders, levies a higher risk of neurological deficits after treatment as a result of the risk of embolization or trauma to the main vessel. It is thus is very helpful to be able to visualize single branches of malformations with multiple feeders, elucidating how exactly the malformation receives flow from feeders. When embolizing malformations, it is important to first embolize pedicles that are the largest and/or feed distal veins and aneurysms. Characteristics of the lesion itself also play a crucial role in dictating embolization approach. For instance, more superficial malformations derive their supply most commonly from peripheral branches of the anterior, middle, and/or posterior cerebral arteries, while larger and deeper malformations may derive their supply from lenticulostriates, choroidals, and similarly located central arterial vessels. The deeper the lesion, the more complex and technically difficult the embolization can potentially be due to overlapping of various vasculature on angiographic imaging alone (Crimmins et al. 2015).
Given the need to exactly ascertain these characteristics prior to embolization, the use of advanced imaging technique and CBCT during embolization of cerebrovascular lesions has exploded in recent years. Lin et al. presented time-resolved CBCT as a technology providing both anatomic and hemodynamic evaluations for real-time classification of AVMs (Lin et al. 2018), and Sandovia-Garcia et al. suggested that this technique could replace conventional CBCT and multiple 2D DSA in diagnostic and therapeutic procedures (Sandoval-Garcia et al. 2017). In their study, Blanc et al. have demonstrated the applicability of CBCT-based 3D-Roadmap for superselective catheterization during AVM embolization (Blanc et al. 2015).
To the best of our knowledge, this retrospective study is the first one to assess a newly released virtual injection software in a context of cerebrovascular lesion diagnosis or treatment. Compared to the other advanced visualization techniques listed above, virtual injection software is designed to minimize CBCT review time and microcatheter manipulations, by simulating selective catheterizations. All superselective catheterizations simulations in this study were evaluated as accurate, even for the more distal segments. This preliminary result suggests that the application of the virtual catherization technology is safe in cerebral CBCTs.
There were minor differences between Reviewer A (15 years of experience) and Reviewer B (5 years of experience). Selective catheterizations simulations seem more useful for less experienced operators, as illustrated by the increased number of POIs selected by Reviewer B and his higher average utility score compared to Reviewer A. Overall however, both neurointerventionalists described virtual catheterization technology as very useful. Potential benefits were identified in procedure planning: identification of the supplying branches, embolization points and minimizing microcatheter manipulations. These observations corroborate the results published for other anatomies, such as trans-arterial liver-directed therapies, where similar technologies have been available for years (Cui et al. 2020). The software was also evaluated as useful for procedure guidance and to segment the targeted vessels in preparation of an overlay to be used with live fluoroscopy. A few studies reported the benefit of augmented fluoroscopy for neurointerventional procedures in increasing operator confidence, reducing radiation dose, and reducing contrast injected into the patient. In these publications, volumes used for 3D-roadmapping were mainly extracted from pre-op imaging (Zhang et al. 2017; Ruijters et al. 2011; Zhang et al. 2016; Kishore et al. 2020), with no time constraint to segment the structures of interest, but inherent risks of misregistration due to change in patient position. In some other studies, the whole vasculature from the CBCT was used as overlay (Blanc et al. 2015; Jang et al. 2016), as careful segmentation of the targeted vessels was too time-consuming to be performed intraoperatively. In this scenario, the interest in using augmented fluoroscopy might be limited due to many vessels overlapping on the 3D roadmap and causing cluttered visualization.
The software has certain limitations. The software is FDA approved for utilization only with angiography and cone-beam CT data acquired on GE hardware. However, the software can be utilized on any single plane and bi-plane GE angiography suite regardless of its age or generation as long as it is a device with cone-beam CT acquisition capability. An artificially wrong course can potentially be mapped if there is significant venous contamination during the cone-beam CT as subjacent opacified venous structures would interfere with the software’s ability to detect clear arterial trajectories. This can also been seen if there is broad or excessive arteriovenous shunting proximal to the malformation’s nidus. As such, the software’s utilization is best exemplified in clinical situations in which navigation towards a distal aneurysm or focal AVM nidus (and potentially an arteriovenous fistula communication point) without excessive surrounding venous contamination proximal to the target is needed.
This study had notable limitations. The retrospective design did not allow us to assess the software performance in real time procedure settings. The small number of patients included is another concern. The small field of view is an intrinsic limitation of such CBCT based vessel detection software, as the software does not track all vessels exiting the CBCT volume. Another common reason given by the reviewers for low utility scores is some patients’ straightforward anatomy, where a quick review of CBCT alone was enough to complete the procedure effectively and efficiently. Additionally, clinical judgment is warranted to select the utility of performing CBCT in a given cerebrovascular malformation, as each patient’s respective anomaly may possess varying degrees of vessel friability and highly suspicious features (such as flow-rated or intra-nidal aneurysms) making them prone to vessel injury ranging from spasm to potential rupture. Such risks may be mitigated with alterations in acquisition technique. These include allowing a rise to achieving final rate of contrast injection, injecting from a common carotid artery rather than direct internal carotid, thereby distributing the total volume of contrast across other surrounding vascular territories (though this may degrade the ability of the software to identify target vessels in the setting of more surrounding opacified vessels).
We believe virtual injection software will promote the adoption of augmented fluoroscopy by enabling rapid segmentation of the targeted vessels. Specifically, this technology adds value via its capability of identifying the appropriate vascular course on 3D images with subsequent allowance for use of the selected route to navigate more efficiently during real-time catheterization and angiography. This software had the potential to be useful not only in the navigation to distal aneurysms and AVM niduses. Precise target vessel selection during catheter directed delivery of chemotherapeutic or immunomodulating agents to intracranial neoplasia is one possible utility of this software, noting that the added reassurance of accurate vessel supply to a given lesion has been of great clinical benefit in the interventional oncology realm of non-neural axis viscera (Cui et al. 2020). Similarly, navigation to branches of the external carotid artery during embolization of head and neck region neoplasia, vascular malformations, or arteriovenous fistula feeders is also technically feasible, noting software limitations that may arise as described earlier.