A Position Statement on Endovascular Models and Effectiveness Metrics for Mechanical Thrombectomy Navigation, on behalf of the Stakeholder Taskforce for AI-assisted Robotic Thrombectomy (START)

Harry Robertshaw; Anna Barnes; Phil Blakelock; Raphael Blanc; Robert Crossley; Rebecca Fahrig; Ameer E. Hassan; Benjamin Jackson

cs.ROMar 30, 2026

A Position Statement on Endovascular Models and Effectiveness Metrics for Mechanical Thrombectomy Navigation, on behalf of the Stakeholder Taskforce for AI-assisted Robotic Thrombectomy (START)

Authors: Harry Robertshaw, Anna Barnes, Phil Blakelock, Raphael Blanc, Robert Crossley, Rebecca Fahrig, Ameer E. Hassan, Benjamin Jackson,

Organizations: School of Biomedical Engineering & Imaging Sciences, Kings College London, London, UK · King’s Technology Evaluation Centre (KiTEC), Kings College London, London, UK · Patient & Public Involvement, Kings College London, London, UK · Department of Interventional Neuroradiology, Rothschild Foundation Hospital, Paris, France · Department of Interventional Neuroradiology, North Bristol NHS Trust, Bristol, UK · Siemens Healthineers AG, Erlangen, Germany · Pattern Recognition Lab, FAU Erlangen-Nürnberg, Erlangen, Germany · Department of Neurology, University of Texas Rio Grande Valley, Edinburg, TX, USA · Society of Vascular and Interventional Neurology, Minneapolis, MN, USA · Department of Artificial Intelligence in Biomedical Engineering, FAU Erlangen-Nürnberg, Erlangen, Germany · Stroke Association, London, UK · European Society of Minimally Invasive Neurological Therapy, Zurich, Switzerland · Department of Neuroradiology, Kings College Hospital, London, UK · Department of Neuroradiology, Queens Hospital Romford, London, UK · Exemplar Devices LLC, Beavercreek, OH, USA · St. Michael’s Hospital, Toronto, Canada · KU Leuven University, Leuven, Belgium · Queen Mary University of London, London, UK · Translational & Clinical Research Institute, Newcastle University, Newcastle, UK · Newcastle upon Tyne Hospitals NHS Foundation Trust, Newcastle, UK · United Kingdom Neurointerventional Group, London, UK

Abstract

While we are making progress in overcoming infectious diseases and cancer; one of the major medical challenges of the mid-21st century will be the rising prevalence of stroke. Large vessels occlusions are especially debilitating, yet effective treatment (needed within hours to achieve best outcomes) remains limited due to geography. One solution for improving timely access to mechanical thrombectomy in geographically diverse populations is the deployment of robotic surgical systems. Artificial intelligence (AI) assistance may enable the upskilling of operators in this emerging therapeutic delivery approach. Our aim was to establish consensus frameworks for developing and validating AI-assisted robots for thrombectomy. Objectives included standardizing effectiveness metrics and defining reference testbeds across in silico, in vitro, ex vivo, and in vivo environments. To achieve this, we convened experts in neurointervention, robotics, data science, health economics, policy, statistics, and patient advocacy. Consensus was built through an incubator day, a Delphi process, and a final Position Statement. We identified that the four essential testbed environments each had distinct validation roles. Realism requirements vary: simpler testbeds should include realistic vessel anatomy compatible with guidewire and catheter use, while standard testbeds should incorporate deformable vessels. More advanced testbeds should include blood flow, pulsatility, and disease features. There are two macro-classes of effectiveness metrics: one for in silico, in vitro, and ex vivo stages focusing on technical navigation, and another for in vivo stages, focused on clinical outcomes. Patient safety is central to this technology's development. One requisite patient safety task needed now is to correlate in vitro measurements to in vivo complications.

Figures & tables

Figure 1: Flowchart for methods followed in this study, including incubator meeting, three rounds of Delphi, and final Position Statement.

Figure 2: Defined phases of MT intervention: (A1) primary access (femoral artery), (A2) primary access (radial artery), (B) secondary access, (C) treatment (stent shown here, but could also be an aspiration catheter) (D) removal of navigation equipment (and access closure). The “navigation phases" are considered to be (A) primary and (B) secondary access.

Figure 3: Consensually agreed benefits and risks of robotic mechanical thrombectomy (MT), both with and without artificial intelligence (across all developmental stages of robotic MT from in silico to in vivo ). Percentage agreement for each benefit and risk was taken from the round where consensus was first reached. Benefits and risks of robotic MT that did not reach consensus in any round can be found in Supplementary Table 1 .

Figure 4: Consensus of which mechanical thrombectomy (MT) phases are effective for the development of robotic MT both with and without artificial intelligence (across all developmental stages of robotic MT from in silico to in vivo ), for testbeds from each developmental stage. Percentage agreement for each phase was taken from the round where consensus was first reached.

Experimental Factor	in silico	in vitro	ex vivo	in vivo (non-human)
Realistic vessel anatomy / Human-like vessel anatomy	84% ( $16/19$ )	95% ( $18/19$ )	84% ( $16/19$ )	95% ( $21/22$ )
Deformable vessels	89% ( $17/19$ )	84% ( $16/19$ )	84% ( $16/19$ )	84% ( $16/19$ )
Simulated blood flow	82% ( $18/22$ )	84% ( $16/19$ )	86% ( $18/21$ )	—
Simulated pulsatility	86% ( $18/21$ )	95% ( $19/20$ )	—	—
Use of both catheter and guidewire for navigation as opposed to single item	95% ( $18/19$ )	95% ( $18/19$ )	89% ( $17/19$ )	89% ( $17/19$ )
Diseased vessel containing atherosclerotic plaques	90% ( $19/21$ )	86% ( $19/22$ )	80% ( $16/20$ )	90% ( $18/20$ )

Table 1: Consensus of which experimental factors were considered important for each developmental stage testbed (both with and without artificial intelligence). Percentage agreement for each factor was taken from the round where consensus was first reached. Note that blood flow and pulsatility would not need to be simulated in vivo . Experimental factors that did not reach consensus in any round can be found in Supplementary Table 2 .

Effectiveness Measure	in silico	in vitro	ex vivo	in vivo
Success rate	100% ( $20/20$ )	100% ( $19/19$ )	100% ( $19/19$ )	—
Number of phases or steps [ 28 ]	84% ( $16/19$ )	90% ( $19/21$ )	84% ( $16/19$ )	—
Number of failures (e.g., wrong branch catheterization)	95% ( $19/20$ )	89% ( $17/19$ )	89% ( $17/19$ )	—
Number of handling errors made [ 28 ]	84% ( $16/19$ )	89% ( $17/19$ )	95% ( $18/19$ )	—
Procedure time	85% ( $17/20$ )	89% ( $17/19$ )	86% ( $19/22$ )	—
Path following error	81% ( $17/21$ )	81% ( $17/21$ )	—	—

Table 2: Consensus on important effectiveness measures for each developmental stage (with and without AI), as well as during in vivo clinical-like assessment. Percentage agreement for each factor was taken from the round where consensus was first reached. Measures that did not reach consensus in any round can be found in Supplementary Table 3 .

Item	in silico	in vitro	ex vivo	in vivo
MT Navigation Phases				(non-human)
Phase A1	✓	✓	✓	✓
Phase A2	✓	✓	✓	✓
Phase B	✓	✓	✓	✓
Phase C	✓	✓	✓	✓
Phase D	—	—	✓	—

Table 3: Combined consensus recommendations for testbeds and effectiveness measures across developmental stages. We have also given units of measure. MT: mechanical thrombectomy; eTICI: extended treatment in cerebral infarction; mTICI: modified treatment in cerebral infarction.

Figure 5: Evolution of testbed complexity with feature complexity. Examples of system complexity—simple in silico , simple in vitro , standard in silico , standard in vitro , standard ex vivo , complex ex vivo , and complex in vivo - are shown in Supplementary Figure 1 through 7 , respectively.

Domain	Benefit/Risk	Agreement
Benefit	To decrease patient radiation exposure	68% ( $15/22$ )
	To decrease patient contrast exposure	41% ( $9/22$ )
	To identify best devices or approach for a particular patient	68% ( $15/22$ )
	To support training of new operators	68% ( $15/22$ )
	To standardize procedures	64% ( $14/22$ )
	To increase instrument dexterity and improve intuitive navigation (reducing the workload of the operator)	73% ( $16/22$ )

List of˙Supplementary Tables 1 Benefits and risks without consensus after three rounds for robotic MT, both with and without AI (across all developmental stages of robotic MT from in silico to in vitro ). Percentage agreement for each benefit and risk was taken after all rounds.

Developmental Stage	Factor	Very Effective/ Somewhat Effective
in silico	Simulated respiration	43% ( $9/21$ )
in vitro	Simulated respiration	35% ( $7/20$ )
ex vivo	Simulated respiration	55% ( $11/20$ )
ex vivo	Simulated blood flow	20% ( $4/20$ )

List of˙Supplementary Tables 2 Non-consensual items after exploring which experimental factors were considered important for each developmental stage testbed (both with and without AI). Percentage agreement for each item was taken after all three rounds.

Developmental Stage	Effectiveness Measure	Very Effective/ Somewhat Effective
in silico	Path length	76% ( $16/21$ )
	Instrument tip speed	76% ( $16/21$ )
	Instrument tip acceleration	67% ( $14/21$ )
	Volume of contrast agent	48% ( $10/21$ )
	Number of guidewire tip touches on the vessel wall	75% ( $15/20$ )
in vitro	Instrument tip acceleration	75% ( $15/20$ )

List of˙Supplementary Tables 3 Non-consensual items after exploring which effectiveness measures were important for each developmental stage (both with and without AI) as well as during in vivo clinical assessment. Percentage agreement for each item was taken after all three rounds.

Category	Key Performance Indicators (KPIs) / Milestones
Reproducible technical KPIs	Device compatibility: Demonstrated use with standard devices (guide/sheath, aspiration, stent-retrievers) across multiple vendors; rapid tool-change workflow.
	Operating room integration and interoperability: Compatible with biplane angiography suites; adequate cyber-security (e.g., ISO 27001/NIST), medical IT-network risk management (IEC 80001-1).
	Teleoperation performance: Adequate end-to-end latency, jitter, link availability; automatic fail-safe modes and “safe stop/withdraw” on dropouts.
Prospective human data	Showing non-inferior morbidity and mortality outcomes.
	Faster patient access.
Financial viability	Confirmed commissioned network pilots: funding/reimbursement including training of staff.

List of˙Supplementary Tables 4 Example proposed technical, clinical, economic, and regulatory milestones for safe translation of robotic endovascular systems.

List of˙Supplementary Figure 1 Example of in silico testbed.

List of˙Supplementary Figure 2 Example of simple in vitro testbed.

List of˙Supplementary Figure 3 Example of standard in silico testbed. Adapted from: Kreiser et al. 2021 [ 52 ] . Reused under the terms of the Creative Commons Attribution 4.0 International License (CC BY 4.0; http://creativecommons.org/licenses/by/4.0/). The image has been cropped and annotated by the authors.

List of˙Supplementary Figure 4 Example of standard in vitro testbed. Adapted from Fischer et al. 2023 [ 44 ] . Reused under the terms of the Creative Commons Attribution 4.0 International License (CC BY 4.0; http://creativecommons.org/licenses/by/4.0/). The image has been cropped and annotated by the authors.

List of˙Supplementary Figure 5 Example of standard ex vivo testbed.

List of˙Supplementary Figure 6 Example of complex ex vivo testbed.

List of˙Supplementary Figure 7 Example of complex in vivo testbed. Adapted from Peng et al. 2023 [ 53 ] . Reused under the terms of the Creative Commons Attribution 4.0 International License (CC BY 4.0; http://creativecommons.org/licenses/by/4.0/). The image has been cropped and annotated by the authors.

Robotic Endovascular Intervention

Robotic endovascular intervention uses robotic systems to navigate vascular instruments and perform procedures with greater precision, control, or autonomy.

See all 24 papers

Jul 10, 2026

Vascular Geometry Characterization for AI-Based Endovascular Navigation

Mechanical thrombectomy (MT) is a time-critical intervention for acute ischemic stroke; however, access remains limited due to a shortage of neuroradiologists and specialized centers. Reinforcement learning (RL) offers potential to automate endovascular navigation and improve accessibility, yet current models lack standardized frameworks to assess navigation difficulty for model training and evaluation. This study aims to identify vascular metrics associated with navigation difficulty and to develop an automated pipeline for quantitative vascular feature extraction, enabling future complexity grading. Vascular trees were segmented from computed tomography angiograms from 61 patients, and vascular metrics including aortic arch type, presence of bovine arch, vessel length, tortuosity, take-off angle, number of reverse curves, were measured using a custom pipeline. A Soft Actor-Critic RL algorithm was used for 120 s autonomous navigation. Outcomes were analyzed using both mixed effects linear and logistic regression. On the left side, the presence of a bovine arch and aortic arch type II/III increased navigation time by 30.19 s and 37.92 s, respectively, while greater tortuosity (\b{eta} = 118.20) further prolonged the procedure and reduced success probability. On the right side, type II/III arches extended procedure time by 45.94 s, while each additional reverse curve was associated with 3.96 s longer navigation time and lower probability of success. These findings demonstrate for the first time that MT agent navigation difficulty is strongly influenced by vascular geometry. The proposed automated pipeline enables objective and quantitative characterization of vascular features, providing a foundation for future development of standardized complexity grading and RL model evaluation, without aiming to demonstrate clinically generalizable autonomous navigation.

Han-Ru Wu, Harry Robertshaw, Lisa Dwyer-Joyce +2

Department of Medical Imaging Science National Taiwan University Hospital Taiwan · Surgical & Interventional Engineering School of Biomedical Engineering & Imaging Sciences Kings College London UK · Chelsea & Westminster Hospital UK

Reinforcement Learning Medical Image Analysis Surgical Robotics

Ask PDF

Feb 20, 2026

Toward AI Autonomous Navigation for Mechanical Thrombectomy using Hierarchical Modular Multi-agent Reinforcement Learning (HM-MARL)

Mechanical thrombectomy (MT) is typically the optimal treatment for acute ischemic stroke involving large vessel occlusions, but access is limited due to geographic and logistical barriers. Reinforcement learning (RL) shows promise in autonomous endovascular navigation, but generalization across 'long' navigation tasks remains challenging. We propose a Hierarchical Modular Multi-Agent Reinforcement Learning (HM-MARL) framework for autonomous two-device navigation in vitro, enabling efficient and generalizable navigation. HM-MARL was developed to autonomously navigate a guide catheter and guidewire from the femoral artery to the internal carotid artery (ICA). A modular multi-agent approach was used to decompose the complex navigation task into specialized subtasks, each trained using Soft Actor-Critic RL. The framework was validated in both in silico and in vitro testbeds to assess generalization and real-world feasibility. In silico, a single-vasculature model achieved 92-100% success rates on individual anatomies, while a multi-vasculature model achieved 56-80% across multiple patient anatomies. In vitro, both HM-MARL models successfully navigated 100% of trials from the femoral artery to the right common carotid artery and 80% to the right ICA but failed on the left-side vessel superhuman challenge due to the anatomy and catheter type used in navigation. This study presents the first demonstration of in vitro autonomous navigation in MT vasculature. While HM-MARL enables generalization across anatomies, the simulation-to-real transition introduces challenges. Future work will refine RL strategies using world models and validate performance on unseen in vitro data, advancing autonomous MT towards clinical translation.

Harry Robertshaw, Nikola Fischer, Lennart Karstensen +6

School of Biomedical Engineering & Imaging Sciences, Kings College London, London, UK · School of Engineering & Materials Science, Queen Mary University of London, London, UK

Robotic Endovascular Intervention Multi-Agent Reinforcement Learning Hierarchical RL

Ask PDF

Apr 22, 2026

Toward Safe Autonomous Robotic Endovascular Interventions using World Models

Autonomous mechanical thrombectomy (MT) presents substantial challenges due to highly variable vascular geometries and the requirements for accurate, real-time control. While reinforcement learning (RL) has emerged as a promising paradigm for the automation of endovascular navigation, existing approaches often show limited robustness when faced with diverse patient anatomies or extended navigation horizons. In this work, we investigate a world-model-based framework for autonomous endovascular navigation built on TD-MPC2, a model-based RL method that integrates planning and learned dynamics. We evaluate a TD-MPC2 agent trained on multiple navigation tasks across hold out patient-specific vasculatures and benchmark its performance against the state-of-the-art Soft Actor-Critic (SAC) algorithm agent. Both approaches are further validated in vitro using patient-specific vascular phantoms under fluoroscopic guidance. In simulation, TD-MPC2 demonstrates a significantly higher mean success rate than SAC (58% vs. 36%, p < 0.001), and mean tip contact forces of 0.15 N, well below the proposed 1.5 N vessel rupture threshold. Mean success rates for TD-MPC2 (68%) were comparable to SAC (60%) in vitro, but TD-MPC2 achieved superior path ratios (p = 0.017) at the cost of longer procedure times (p < 0.001). Together, these results provide the first demonstration of autonomous MT navigation validated across both hold out in silico data and fluoroscopy-guided in vitro experiments, highlighting the promise of world models for safe and generalizable AI-assisted endovascular interventions.

Harry Robertshaw, Nikola Fischer, Han-Ru Wu +6

Surgical & Interventional Engineering, School of Biomedical Engineering & Imaging Sciences, King’s College London, UK · Department of Radiology, National Taiwan University Hospital, Taiwan · Department of Neuroradiology, King’s College Hospital, UK

World Models for Robotics Robotic Endovascular Intervention Surgical Robotics Model-Based RL

Ask PDF

Abstract

Robotic Endovascular Intervention

Explore similar work

Vascular Geometry Characterization for AI-Based Endovascular Navigation

Toward AI Autonomous Navigation for Mechanical Thrombectomy using Hierarchical Modular Multi-agent Reinforcement Learning (HM-MARL)

Toward Safe Autonomous Robotic Endovascular Interventions using World Models