Manual, Joystick, or Haptic Control? An In Vitro Comparison of Navigation Strategies for Robotic Interventional Neuroradiology Procedures

Technical Summary: Manual, Joystick, or Haptic Control? An In Vitro Comparison of Navigation Strategies for Robotic Interventional Neuroradiology Procedures

Problem Statement
Mechanical thrombectomy (MT) is the standard treatment for large-vessel occlusion strokes, yet its efficacy is limited by procedural complexity and a shortage of trained interventional neuroradiologists. Robotic systems offer a potential solution through increased precision and teleoperation capabilities. However, a "translational gap" exists between the intuitive manual manipulation of devices in the operating room and the use of robotic interfaces. While robotic interfaces generally utilize either non-isomorphic inputs (e.g., joysticks) or device-mimicking interfaces (replicating clinical interactions), there is a lack of physical validation comparing these modalities. Furthermore, the integration of haptic feedback to replicate tactile conditions in MT robotics remains a developing field. Crucially, assessing the safety of these interfaces requires quantifying vessel-wall forces, which is challenging in traditional physical phantoms that lack objective force sensing.

Methodology
The study employed a custom-built in vitro experimental platform consisting of three core components:

1. Robotic Controller: A device-mimicking interface adapted from previous work, capable of translation and rotation. It operated in two modes: with haptic feedback enabled (closed-loop system using distal sensors to render resistive forces at 250 Hz) and with haptic feedback disabled.
2. Endovascular Robot: A custom robot utilizing Nema 17 stepper motors and lead screw mechanisms to provide two degrees of freedom (translation and rotation) for concentric neuroendovascular devices. It integrated an ATI Mini 58 six-DoF force sensor to measure interaction forces for haptic rendering.
3. Sensorized Neurovascular Phantom: A 3D-printed vascular phantom fabricated from flexible photopolymer resin, designed from anonymized patient imaging data. It featured eight embedded piezoresistive force sensors (based on Velostat film) to measure local device-wall contact forces. The phantom was submerged in glycerin with external radiopaque markers to simulate fluoroscopic visualization.

Experimental Design
Ten interventional neuroradiologists (4 novices, <5 years experience; 6 experts, >5 years experience) performed simulated navigations through a tortuous path (descending aorta to right internal carotid artery). Each participant completed eight trials using four control modalities:

1. Manual navigation (reference).
2. Robotic controller with haptic feedback (device-mimicking).
3. Robotic controller without haptic feedback (device-mimicking).
4. Joystick input (non-device-mimicking baseline).

Key Contributions

1. Sensorized Phantom Development: The authors developed a modular, 3D-printed vascular phantom with integrated piezoresistive sensors, enabling the quantitative assessment of device tip forces during robotic navigation in a repeatable, controlled environment.
2. Comparative Verification: The study compared the performance of robotic controllers (with and without haptics) against manual control and joystick interfaces across multiple metrics.
3. Experience-Based Analysis: The impact of operator experience (novice vs. expert) was assessed across all control modalities to determine how seniority influences performance and safety.

Results

• Navigation Time: Manual navigation was significantly faster (mean 47.7 s) than all robotic modalities (Haptic On: 248.7 s; Haptic Off: 314.7 s; Joystick: 392.6 s; p<0.001). Among robotic controllers, haptic feedback showed a trend toward faster times than haptic-off or joystick, though this was not statistically significant.
• Safety (Forces): All modalities were deemed safe, with maximum vessel-wall forces remaining well below the conservative 0.70 N puncture threshold. Manual navigation produced the lowest forces. Robotic controllers produced slightly higher forces, but differences were not statistically significant.
• Errors:
  • Incorrect Catheterizations: Manual navigation had the lowest error rate. Experts made significantly fewer errors than novices (0.25 vs. 0.62; p=0.035) across all controllers.
  • Prolapse Events: Joystick navigation resulted in significantly more prolapse events (1.56) compared to manual control (0.13; p=0.018). Haptic feedback showed a trend toward reducing prolapse compared to haptic-off, but this was not statistically significant.
• User Feedback: Manual navigation was rated as the most intuitive. Among robotic interfaces, device-mimicking controllers (both haptic on and off) were perceived as significantly more intuitive than the joystick (p=0.033). Participants generally preferred device-mimicking controllers over joysticks.

Significance and Claims
The paper concludes that device-mimicking robotic controllers outperform joystick interfaces on most metrics, including speed, safety (fewer prolapse events), and user intuitiveness. While manual navigation remains the fastest and most accurate, the authors argue that the additional time required for robotic navigation is negligible when compared to the transfer times to hospitals capable of performing mechanical thrombectomy.

The study claims that robotic interfaces are safe within the tested paradigm, as all forces remained orders of magnitude below puncture thresholds. Haptic feedback shows "promising but non-significant" performance benefits, particularly in reducing the performance gap between novices and experts and potentially lowering prolapse rates, though the authors note these findings require further validation. The work establishes a framework for the controlled evaluation of controller interfaces and feedback modalities, suggesting that device-mimicking control is a viable and preferred approach for robotic neurointervention, potentially bridging the translational gap between manual and robotic practice.