2025 Volume 19 Issue 1 Article ID: oa.2025-0028
Objective: Artificial intelligence (AI) holds promise for advancing neuroendovascular therapy through device evaluation, but its application in real-world clinical settings remains limited. We aimed to validate the feasibility of AI-based quantitative device evaluation during actual procedures by assessing the stability of the Rist radial access guide catheter (Medtronic, Dublin, Ireland), a novel device designed for the increasingly adopted transradial approach (TRA), during flow diverter stent (FDS) placement.
Methods: We retrospectively analyzed 4 cases of FDS placement using Rist via the TRA. Rist was tracked in recorded fluoroscopic videos using the AI technology of Neuro-Vascular Assist (iMed Technologies, Tokyo, Japan). The movement distance of Rist during FDS placement was calculated as a stability indicator.
Results: All procedures were successfully completed without any complications. Rist was introduced from the radial artery and positioned in the distal internal carotid artery. The maximum movement distances of the Rist during the procedures were 3.36, 6.63, 1.79, and 0.33 mm for each case, respectively, with an average of 3.03 mm. The maximum movement distances per minute were 1.68, 2.34, 1.19, and 0.46 mm/min, respectively, with a mean of 1.42 mm/min. These measurements suggest sufficient stability for the FDS procedures.
Conclusion: This study demonstrates the feasibility of using AI technology to quantitatively analyze Rist stability in TRA procedures. To the best of our knowledge, this is the 1st clinical evaluation of device function in a clinical setting using AI technology. Further studies with more cases are required to validate these findings. This method is promising for real-world device evaluation and development.
artificial intelligence
FDSflow-diverter stent
FNfalse negative
GCguide catheters
ICAinternal carotid artery
TFAtransfemoral approach
TRAtransradial approach
Artificial intelligence (AI) has shown significant potential in advancing neuroendovascular therapy. In particular, AI has been applied to improve the recognition accuracy of devices such as guidewires and guide catheter (GC) tips in fluoroscopic images and to track device movements.1–5) However, most studies have relied on phantoms or simulators, and research focusing on real-world evaluation of devices in actual clinical cases is limited. Quantitative evaluation of device movements could provide valuable insights into operator proficiency and procedural difficulty, yet these methods remain underexplored in clinical settings.
Neuro-Vascular Assist (iMed Technologies, Tokyo, Japan) is an AI-based software approved for real-time support during neuroendovascular procedures in Japan. Previous studies have explored its potential clinical utility in providing intraoperative real-time support during carotid artery stenting, coil embolization, liquid embolization, and mechanical thrombectomy for acute ischemic stroke.6–10) These studies focused on AI notifications for movements of devices such as guidewire, GC, and filter. However, they did not quantitatively evaluate device performance.
Recently, the transradial approach (TRA) has been adopted in neuroendovascular therapy.11–13) However, using conventional GCs via the TRA is often challenging, particularly during the turn into the great vessels from the arm, making navigation to the target vessel more difficult compared with the transfemoral approach (TFA).14) To address this issue, the Rist (Medtronic, Dublin, Ireland) was developed, featuring enhanced flexibility and hydrophilic coating to improve navigability.14) The Rist is the 1st TRA-specific GC and has been reported to have a lower crossover rate from TRA to TFA compared with conventional GCs.14) However, enhancing flexibility may compromise stability. Although GC stability is a critical factor influencing device performance, objective and quantitative methods for evaluating stability are lacking.
This study aimed to explore the feasibility of using AI technology to quantitatively evaluate GC stability in real-world clinical settings. Using the AI-embedded Neuro-Vascular Assist, we retrospectively analyzed surgical recorded videos to track device movements and assess whether this approach could be applied to evaluate device performance. For this proof of concept, we focused on flow diverter stent (FDS) placement via TRA, which is commonly performed in neuroendovascular therapy for intracranial aneurysms and requires substantial catheter support owing to the use of large devices.15)
This study was approved by the institutional ethics committee (Approval No. 2024-7), and the requirement for informed consent was waived because of the retrospective nature of this study.
From February to May 2024, 7 consecutive cases of FDS placement via TRA using Rist were performed at Mishuku Hospital. At our institution, computed tomography (CT) angiography of the aorta and neck is performed prior to catheterization. The data were transferred to a 3-dimensional (3D) workstation and reconstructed for measurement. The reconstructed CT angiogram and arch aortogram were used to evaluate the aortic arch type and vessel tortuosity. The classification of aortic arch types was based on a method in which a tangent (index line) is drawn at the top of the arch, with reference lines established below it. Specifically, a 1st reference line is drawn at a distance equal to the diameter of the common carotid artery (CCA) below the index line, and a 2nd reference line is drawn further inferiorly. If all vessels (the brachiocephalic artery, left common carotid artery, and left subclavian artery) branch between the index line and the 1st reference line, the arch is classified as Type I; if the branches occur between the 1st and 2nd reference lines, it is classified as Type II; and if the branching extends below the 2nd reference line, it is classified as Type III. A bovine arch is defined as either a common origin of the innominate artery and the left CCA or as a left CCA that originates directly from the innominate artery.16)
Rist radial access system17)The Rist radial access system (Medtronic) is the 1st GC designed specifically for TRA in neuroendovascular therapy. It comprises a 5.5-Fr selective Sim2 catheter and a 0.079-in. (7-Fr) Rist GC, available in lengths of 95 cm, 100 cm, and 105 cm. For this study, only the 95-cm Rist GC was used. The Rist features optimized stiffness to enhance proximal stability and distal navigability, facilitating its use in the radial approach. Its distal flexible segment measures 29.5 cm, with a hydrophilic coating over 25 cm of this segment to further improve navigability.
Interventional proceduresAll procedures were performed by a single operator (S.T.) under general anesthesia, using a single-plane system (Innova IGS 530, GE Healthcare, Chicago, IL, USA). The same system was used in all cases. During these procedures, no real-time AI assistance was used, and operator decisions and catheter manipulations were based solely on standard clinical practice. Conventional or distal radial access was established under ultrasound guidance. Except for Cases 1 and 6, a 7-Fr Glidesheath Slender (Terumo, Tokyo, Japan) was used in all procedures. In contrast, for Cases 1 and 6, the tip of a 7-Fr Peel-off Sheath (Medikit, Tokyo, Japan) was initially inserted, and following a Rist insertion, the sheath was removed so that the procedure was performed sheathless. A 7-Fr Rist was used and advanced to the target vessel with the assistance of a selective Sim2 catheter (Medtronic). The GC was positioned as distally as possible within the target vessels. A 5-Fr Navien (Medtronic) was used as the distal support catheter. A Phenom 27 (Medtronic) was navigated distally to the aneurysm. A Pipeline Shield (Medtronic) was also deployed. Since a single-plane system was used, the angle of the flat panel was adjusted as needed during FDS deployment. A percutaneous transluminal angioplasty was performed. Recordings of fluoroscopic images were made using subtracted images when available or using unsubtracted images otherwise.
Neuro-Vascular Assist (an AI-based system)Neuro-Vascular Assist is an AI-based system developed to provide real-time support during neuroendovascular procedures.6–9) The system analyzes fluoroscopic images in real-time, tracking devices. The system notifies operators about movements of devices such as filters, wires, and GCs. While primarily designed for intra-procedural assistance, our study used its device tracking system retrospectively to analyze fluoroscopic videos for quantitative device evaluation.
Retrospective analysis by AI using recorded videosThe analysis by AI was performed on recorded fluoroscopic videos after all procedures were completed. The analysis was performed on the longest segment without screen movement or magnification changes during FDS deployment using the AI technology of the Neuro-Vascular Assist to track the tips of the Rist and Navien catheters (Fig. 1, Supplementary Video). The conversion from pixel values to actual measurements was based on the diameter of the Rist, which was 0.093 in. The coordinates of the device tips (catheter tips) on the screen were output for each frame of the surgical video (every 0.1 s), and the movement distance on the plane was calculated. The accuracy of the device recognition in all frames was verified by a neuroendovascular specialist (Y.S.).
As described in the “interventional procedures” section, the operator moves the flat panel as needed during FDS deployment, but the angulation changes were sometimes unsuitable for our movement distance analysis. Therefore, 3 cases were excluded. Finally, we selected 4 cases in which the fluoroscopic view without screen movement or magnification was changed for more than half of the FDS deployment time. Patients’ baseline characteristics, including those of excluded cases, are shown in Table 1. Case 1 exhibited a type Ⅱ and bovine-type aortic arch with a left internal carotid artery (ICA) aneurysm, while Case 4 had a type Ⅲ aortic arch with a right ICA aneurysm. None of the cases had double subclavian-innominate curves, proximal carotid loop, or proximal radial loop—factors previously reported to contribute to difficult Rist catheter access.18) The absence of these anatomical challenges facilitated smoother navigation to the target vessel.
| Case | Age, sex | Aortic arch | Puncture site | Aneurysm | |
|---|---|---|---|---|---|
| Location | Maximum diameter (mm) | ||||
| 1 | 50 s, F | Type II/ Bovine | rt. dTRA | lt. ICA C2 portion | 8.5 |
| 2 | 70 s, M | Type II | rt. dTRA | lt. ICA-PcomA | 10.0 |
| 3 | 40 s, F | Type I | rt. cTRA | lt. ICA C2 portion | 7.7 |
| 4 | 70 s, F | Type III | rt. cTRA | rt. ICA-PcomA | 5.6 |
| 5* | 70 s, F | Type III | rt. cTRA | rt. ICA-PcomA | 5.7 |
| 6* | 50 s, F | Type I | rt. dTRA | lt. ICA C2 portion | 6.6 |
| 7* | 40, F | Type I | lt. dTRA | rt. ICA C2 portion | 5.4 |
*excluded case.
cTRA, conventional transradial artery; dTRA, distal transradial artery; ICA, internal carotid artery; lt, left; PcomA, posterior communicating artery; rt, right
In all cases, the Rist was successfully positioned distal to the cervical portion of the ICA, and the Navien was placed in either the petrous or cavernous portion, achieving a 100% catheter placement success rate. Moreover, the FDS was fully deployed, and final angiograms confirmed appropriate stent positioning and vessel wall apposition in all cases, ensuring procedural success without any procedure-related complications. Only 1 Pipeline Shield stent (Medtronic) was used per procedure. The Rist provided sufficient stability, requiring no additional operator adjustments during FDS deployment and enabling smooth manipulation of the Navien for precise stent placement. The operator did not perceive any instability of the Rist during these procedures. Procedural timing metrics for each case are, including those of excluded cases, shown in Table 2. They indicate that the entire process was smoothly preformed.
| Case | Puncture to target vessel time (min) |
FDS deployment Time (min) |
Fluoroscopy time (min) |
Fluoroscopy dose (mGy) |
Total procedure time (min) |
|---|---|---|---|---|---|
| 1 | 6 | 5 | 41 | 2025 | 58 |
| 2 | 4 | 7 | 37 | 1538 | 71 |
| 3 | 3 | 11 | 46 | 2514 | 115 |
| 4 | 4 | 10 | 21 | 1245 | 51 |
| 5* | 4 | 10 | 39 | 1937 | 86 |
| 6* | 4 | 10 | 30 | 1689 | 59 |
| 7* | 6 | 7 | 36 | 3036 | 88 |
*excluded case.
FDS, flow diverter stent
We performed analyses for 120, 170, 90, 40 s in the 4 cases. These durations corresponded to periods in which the fluoroscopic view remained stable without screen movement or magnification changes for more than half of the FDS deployment time, as mentioned earlier. The operator (ST) reviewed the original videos during the analysis periods and confirmed that there was no intentional movement of the Rist during these intervals. This implies that any movement of the Rist during these periods was due to the manipulation of other devices, and it was considered desirable for the Rist to remain stable without movement during FDS deployment.
The maximum movement distances of the Rist during the procedures were 3.36, 6.63, 1.79, and 0.33 mm for each case, respectively, with an average of 3.03 mm. The maximum movement distances per minute were 1.68, 2.34, 1.19, and 0.46 mm/min, respectively, with a mean of 1.42 mm/min (Figs. 2 and 3, Supplementary Video). The maximum movement distance was calculated as the difference between the maximum and minimum distances from the initial position of each device (Fig. 2E–2H). In Case 2, we observed a slight catheter displacement at approximately 120 s (Fig. 2B), which occurred while pushing the FDS for deployment.
There were no false positives in the device detection, and no manual corrections were required. The device recognition rate was 95.8% (4044/4220 frames), with 4.2% (176 frames) false negatives (FNs). The longest FN interval was 49 frames (4.9 s), during which no movement of the Rist was visually observed. Other FNs showed no visible movement of the Rist. Therefore, we concluded that these FNs did not affect the measurement of the Rist’s movement distance.
In this study, we used AI technology to evaluate devices in actual clinical cases. Specifically, in 4 cases of FDS deployment via the TRA, AI automatically measured the movement of the Rist tip, with an average maximum movement per minute of 1.42 mm/min. This method shows potential as a new technique for objective, quantitative, and automated evaluation of GC stability. The use of recorded videos makes it possible to retrospectively analyze device evaluation, providing greater flexibility compared with previous real-time use of the AI system, Neuro-Vascular Assist, during procedures.6–9) To the best of our knowledge, this is the 1st study to quantitatively evaluate these devices in a real-world clinical setting.
Evaluation of guide catheter stability using AIAlthough there are several studies on AI using phantoms or simulators,1–5) a quantitative evaluation of GC stability in real-world clinical settings has not been performed. This is likely due to insufficient technology for tracking devices in clinical X-ray images. Neuro-Vascular Assist, which is now approved in Japan, is being used for real-time support in clinical practice. This includes functions such as tracking the GC and notifying when it moves out of view.6) In this study, we quantitatively evaluated GC stability in real-world settings using AI technology. The recognition rate was 95.8%, and although visual confirmation was necessary, no manual corrections were required, providing sufficient accuracy for device evaluation.
Conventional evaluations of GC stability were mainly based on subjective evaluations such as operator's perception or extreme indicators such as replacing the GC, and detailed quantitative evaluations have not been conducted. For example, previous descriptions regarding Rist have been limited to statements such as, “If Rist was positioned at the level of the carotid bifurcation or lower and no distal access catheter was used, this was often associated with some instability issues”17) and “When proximal support was needed, we were able to advance the Rist GC into the petrous and cavernous segments of the internal carotid artery”,19) without objective evaluation of stability or instability. Rautio et al. classified stability as good, intermediate, and poor, but did not clearly define these categories.17)
Using conventional evaluation methods, these cases would likely be deemed “stable” as the procedures were smoothly completed. The significance of our AI-based GC stability evaluation method lies in its ability to objectively assess these subjective elements. For example, Case 2 showed a maximum movement distance of 6.63 mm and a movement of 2.34 mm per minute, whereas Case 3 had a maximum movement of 1.79 mm and 1.19 mm per minute, visualizing different situations within cases considered “stable” (Fig. 2). In all 4 cases, the maximum movement distance was ≤6.63 mm, and the maximum movement per minute was ≤2.34 mm. These findings suggest that this degree of GC movement does not hinder “sufficient” stability, as the procedures were successful and the operator did not perceive any instability. In addition, using AI-based analysis enabled detailed temporal measurements, which would have been impossible to obtain manually from fluoroscopic videos.
In the future, with more cases, it may be possible to establish criteria for the maximum movement distance or the movement rate per minute, thereby enabling a stability scoring system. This can contribute to device evaluation, exploration of factors affecting stability or instability, assessment of operator skills, and optimization of device combinations.
In this study, we focused on evaluating Navien and Rist, with a particular focus on Rist. As case numbers increase, expanding this analysis to include a broader range of TRA devices will provide a more comprehensive comparison of device stability across various clinical scenarios, which would further refine our understanding of GC stability. Such comprehensive evaluations will contribute to device selection and inform future device development, ultimately improving patient outcomes in neuroendovascular procedures.
Transradial approach and GC stabilityCurrently, there is an increasing trend of using TRA for neuroendovascular treatments.17) Reports suggest that TRA can reduce the puncture site and overall complications.20) Although TRA has the advantage of fewer puncture site complications, the smaller vessel diameter at the puncture site requires the use of GCs with smaller diameters than those used for TFA.14) Attempts have been made to use sheathless balloon-guiding catheters for TRA.21–23) Furthermore, conventional GCs often struggle with “the turn up into the great vessels from the arm”,14) necessitating GCs with higher navigability and stability. Recently, GCs for TRA other than the Rist, such as the Zoom RDL24) and Armadillo catheter,25) have been developed. The Zoom RDL is an 8-Fr radial access system. In a case series by Morsi et al.,24) the incidence of access site complications was reported as 0%, with a 6.9% conversion rate to TFA. While this GC is designed to reach distal vessels such as the M1 segment of the middle cerebral artery, its flexible distal portion makes it less stable when targeting proximal vessels such as the external carotid artery or vertebral artery. However, these findings are based on the operator’s subjective evaluation. The Armadillo catheter is a GC optimized for dual modes: “tracking mode” and “support mode.” In a case series by El Naamani et al.,25) the access site complication rate was 0%, and the conversion rate to TFA was 1.4%. The “support mode” reportedly enhances GC stability; however, a detailed evaluation of the stability provided was not performed. Each of these GCs, including the Rist, has distinct characteristics suited to different procedural needs. Therefore, an objective evaluation of GC performance has become increasingly important to guide their appropriate use and optimize procedural outcomes. By utilizing the AI technology used in this study, it is expected that device evaluations in real-world settings will become more widespread, leading to further advancements.
Potential for device developmentThe balance between the flexibility, rigidity, and coating of GCs is crucial for their navigability and stability.14,17,19) Additionally, patient-specific factors, such as aortic arch shape, tortuosity of the approach route, and arterial sclerosis, contribute to many variations.26) Although current vascular simulation models use silicon models, it is challenging to reproduce all patient factors. Moreover, it is difficult to fully replicate biological conditions, such as friction, hemodynamics, hemostasis, thrombus formation, vessel fragility, and vasospasm.27) Therefore, real-world evaluation is highly important for assessing the navigability and stability of GCs.
To date, most TRA procedures in neuroendovascular treatments have been performed using GCs designed for TFA,14,19) limiting the knowledge specific to TRA. Consequently, a quantitative evaluation in real-world settings is likely to provide objective indicators for the development of TRA-specific GCs. This technology is expected to play a crucial role in the development of future devices.
LimitationsThis study has some limitations. First, the small sample size of only 4 cases is a notable limitation. However, the primary aim of this study was to verify whether quantitative device evaluation using AI technology is feasible in actual cases (a feasibility study). We believe that this objective was achieved in the 4 cases. In the future, we plan to increase the number of cases to explore the possibility of providing feedback in clinical practice.
Second, the use of a single-plane system compromised analysis accuracy. Although 3 out of 7 cases were excluded, the patient and procedural characteristics were comparable between the included and excluded cases, and all procedures were successfully completed without significant complications. Therefore, we believe that these exclusions do not reflect a patient selection bias but are instead attributable to reduced accuracy resulting from adjustments to the angle of the single-plane flat panel during FDS deployment. In addition, measurements were based on 2-dimensional X-ray images rather than 3D images, which may have led to an underestimation of movement distances. However, particularly for the Rist, the target segment of the ICA was relatively straight, as shown in Fig. 2, minimizing the potential impact of projection angles on our measurements. Future studies using biplane imaging or 3D reconstruction techniques could provide additional insights into catheter movements in 3D.
Third, this study did not include comparisons with other GCs or multiple operators, which limits the generalizability of our findings. However, conventional operators’ subjective assessments have deemed the Rist 7-Fr to be stable in this study. To quantitatively express that subjective evaluation, we measured both the maximum movement distance and the maximum movement per minute. At least for the Rist 7-Fr, the results suggest that stability may be achieved if the maximum movement distance is ≤6.63 mm (mean 3.03 mm) and the maximum movement per minute is ≤2.34 mm/min (mean 1.42 mm/min). Future research involving different GCs and operators with varying experience levels can provide more comprehensive insights into device effectiveness.
In this study, using AI technology, we successfully measured the distance moved by the 1st tip during FDS deployment in clinical cases. This suggests for the 1st time the possibility of quantitatively evaluating GC stability in clinical settings. In the future, we anticipate that, by increasing the number of cases, we will be able to establish stability criteria and conduct device comparisons.
This video demonstrates the analysis of 2 cases using artificial intelligence technology. The catheter tips are tracked with squares, and the wire tip is tracked with a triangle. By visualizing the trajectories and temporal position changes of both the Navien and Rist catheters during flow-diverter stent deployment, the video illustrates the quantitative evaluation of the movement of the Navien and Rist catheters.
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
Data availabilityThe data that support the findings of this study are available upon request.
Disclosure statementShunsuke Tanoue had no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. Yuya Sakakura received honoraria from iMed Technologies. Kenichi Kono is the CEO of and holds shares in iMed Technology.
