MRI-Actuated Robotic Catheter

• MRI-actuated robotic catheter is a continuum robotic device employing MRI-compatible actuation techniques (tendon-driven and magnetic) for minimally invasive interventions.
• Key methodologies include tendon-driven designs with lead-screw mechanisms and magnetically actuated microcoils that deliver sub-millimeter tip accuracy.
• Real-time feedback from MR-compatible sensors and advanced control algorithms enables closed-loop control and dynamic compensation in clinical settings.

An MRI-actuated robotic catheter is a continuum robotic device designed to operate within the strong magnetic fields and spatial constraints of a clinical magnetic resonance imaging (MRI) environment. These systems leverage MRI-compatible actuation mechanisms, such as piezoelectric motors or magnetic torques generated by microcoils, to achieve dexterous, closed-loop manipulation of clinical catheters and their guiding sheaths for interventions such as radiofrequency ablation of cardiac tissue. The underlying kinematics, actuation strategies, and control algorithms must account for both electromagnetic and mechanical compatibility, and real-time feedback from positional sensors, including MR-tracking coils or external electromagnetic trackers, ensures accurate navigation inside the patient. MRI-actuated robotic catheters have demonstrated path-following errors well below clinical tolerances, with strong prospects for MR-guided electrophysiology and interventional applications (Wang et al., 2023, Hao et al., 28 Dec 2025).

1. System Architectures: Concentric Tendon-Driven and Magnetically Actuated Designs

MRI-actuated robotic catheters fall into two principal design categories: tendon-driven actuators and magnetically steered continuum robots. The concentric tendon-driven structure, as implemented by the compact MR-conditional module in (Wang et al., 2023), comprises two continuum segments—a guiding outer sheath and an ablation catheter—each offering three degrees of freedom: bending, axial rotation, and translation. These segments are realized using steel or Nitinol backbones, polymer sheaths, and rigid guiding tubes. Almost all custom structural components (transmission modules, adapters, latches) are fabricated from acrylonitrile butadiene styrene (ABS) or Tough 2000 resin, with fasteners in Nylon, while all bearings and shafts utilize ceramic or glass-ball-filled Nylon to eliminate ferromagnetic content.

In contrast, the magnetically actuated robotic catheter (Hao et al., 28 Dec 2025) incorporates current-carrying microcoils along the catheter body. The coils, energized by external current drivers, interact with the constant magnetic field of the MRI scanner to generate torques, producing distributed moments along the continuum that steer the catheter tip. The physical construction must satisfy MR-conditionality (i.e., zero ferromagnetic material) and must route electrical signals using RF-shielded cables to minimize electromagnetic interference, in compliance with ASTM F2503 standards.

2. Actuation and Transmission Mechanisms

Tendon-driven modules route clinical handle inputs to the distal tip via complex linkage assemblies. For the outer sheath, a lead-screw mechanism transmits coaxial knob motion linearly for bending, while the catheter’s pulley-based routing delivers off-axis rotations for tip articulation. Tendons transit from the handle bore through 3D-printed adaption modules: splined couplers propagate rotation for axial spin, and cascaded lead-screw and rack-and-pinion elements reshape carrier rotations into required knob actuations.

Actuation is performed by non-magnetic piezoelectric motors (e.g., PiezoMotor LR23-50), with output shafts generating translation through Nylon lead-screws or rotation via square shafts and plastic gears. All cabling is enveloped with RF-shielded braided sleeving and ferrite filters, suppressing electromagnetic interference. The complete robotic envelope is typically 108 mm × 100 mm × 865 mm (W×H×L) and mounts on a low-profile acrylic base outside the MR bore, with only guiding tubes entering the imaging region.

For magnetically actuated designs, microcoils receive direct current inputs. The local magnetic torque at each coil is given by $\tau_i = m_i \times B_i$, with $m_i = N_i A_i z^c_i$ denoting the coil’s magnetic moment and $B_i = R_i^T B_s$ the field in the coil’s frame. Torque is imparted discretely along the catheter body, steering segments while satisfying MR compatibility.

3. Kinematic Modeling: Constant Curvature and Cosserat-Rod Formulations

Tendon-driven robotic catheters typically employ the constant curvature model for continuum segment kinematics. For a segment of arc length $L$, curvature radius $R$, and bending angle $\phi = L/R$, the tip position $\mathbf{p}$ is expressed as: $$x(\phi,R,\delta) = R (1-\cos\phi)\cos\delta,\quad y(\phi,R,\delta) = R(1-\cos\phi)\sin\delta,\quad z(\phi,R,\delta) = R\,\sin\phi,$$ where $\delta$ is the orientation of the bending plane. Uniform curvature is assumed, neglecting torsion and variable curvature.

The velocity at the tip due to shape rates $\dot\phi, \dot\delta$ is given by: $v = J_\phi \dot\phi + J_\delta \dot\delta,$ with $J_\phi$ and $J_\delta$ defined by partial derivatives of the tip position function. Tendon displacements are mapped to curvature changes via $$\Delta l \approx R \phi,\quad \dot\phi = \frac{\dot l}{R}$$ and motor rotation parameters.

For magnetically actuated catheters, the static Cosserat-rod theory provides a higher-fidelity kinematic model. The center-line position $p(s)$, local orientation $R(s)$, and strain vector $\xi(s)$ are governed by the ODE system: $$p'(s) = R(s) e_3,\quad R'(s) = R(s) \widehat{\xi}(s),\quad \xi'(s) = - K^{-1}(s)[\cdots].$$ The equilibrium equations model distributed force and moment densities, constitutive relations relate internal moments to curvature, and boundary conditions (vanishing tip moment) are solved via shooting methods. Each actuator (coil) applies torque at discrete locations, which is incorporated algebraically at those points (Hao et al., 28 Dec 2025).

4. Feedback Control and Real-Time Compensation

Closed-loop position control is achieved using MR-compatible sensors (typically EM tracking coils or future MR-tracking coils), providing real-time tip position $\bar p$ and orientation $\bar R$. The control law is typically a resolved-rate PD controller in task space: $$\Delta q = K_p J^\dagger (p_{\mathrm{des}} - \bar p) + K_d J^\dagger (\dot p_{\mathrm{des}} - \dot{\bar p}),$$ where $J^\dagger$ is the damped pseudoinverse of the current Jacobian, and gain matrices $K_p, K_d$ modulate positional and velocity errors. An online shape-fitting routine minimally estimates true model parameters $(\phi, \delta)$ by minimizing discrepancy between sensor poses and model predictions: $$\min_{\phi, \delta} \| \bar p - p(\phi, \delta) \|^2 + \lambda \| \mathrm{Log}(\bar R^T R(\phi, \delta)) \|^2.$$ This “model-in-the-loop” adaptation recalibrates the Jacobian at each control step to compensate for frictional losses, tendon hysteresis, and handle backlash.

For magnetically actuated robots, tip position feedback from real-time MR imaging or external cameras is processed to update desired trajectory. Analytical Jacobians computed via the virtual angular-velocity method allow current updates to be found in under 4.1 ms per control loop, orders of magnitude faster than finite-difference approaches (≈95 ms). Closed-loop integration relies on rapid MR feedback (≤10–15 ms latency), keeping total control latency below typical MRI frame rates (Hao et al., 28 Dec 2025).

5. MRI Compatibility: Materials, Electromagnetic, and Mechanical Integration

MRI-conditionality is achieved through exclusive use of non-conductive, non-magnetic structural materials (ABS, Tough resin, Acrylic, Nylon), ceramic or glass-ball-filled bearings, and piezoelectric motors with no ferromagnetic cores or permanent magnets. Cable shielding and filtration ensure compliance with MR safety labeling. Certification testing under clinical imaging sequences (1.5 T scanner) verifies absence of:

• Displacement and torque artifacts
• Image distortion due to electromagnetic interference
• Metallic objects within MR field

Mechanically, the system is constructed to mount unobtrusively on patient infrastructure outside the bore, with only guiding tubes or catheters inside the imaging region. Fast, swappable fixtures enable rapid exchange of components without disrupting scanner setup.

6. Experimental Validation and Performance Metrics

Bench-top validation protocols focus on path-following accuracy. For example, in (Wang et al., 2023), the robot was tasked to follow a 30 mm radius circular trajectory, emulating pulmonary vein ablation, discretized into 72 targets and reached in up to 20 control cycles (1 mm convergence threshold). Comparative evaluation was performed among three schemes: open-loop (no feedback), closed-loop without online fitting, and closed-loop with adaptive (online shape fitting) Jacobian.

The following table summarizes the reported errors:

Control Scheme	Avg Perpendicular Error	Avg In-plane Error	Overall Tip Error
Open-loop (no feedback)	1.45 mm	11.08 mm	> 2 mm
Closed-loop, fixed Jacobian	0.43 mm	0.65 mm	< 2 mm
Closed-loop, adaptive Jacobian	0.50 mm	0.98 mm	< 0.8 mm

Across all segments, closed-loop errors remained nearly constant (<1 mm), well under clinical tolerance. Open-loop error grew linearly with bending angle and could exceed 15 mm for large deflections. Minor spikes at segment transitions were attributed to solver convergence in shape fitting (≤3 mm). In the magnetically actuated system (Hao et al., 28 Dec 2025), root-mean-square positional open-loop errors ranged from 3.7–6.5 mm, but repeatability after offset alignment was <1 mm.

7. Limitations, Future Directions, and Clinical Prospects

Current limitations include occasional nonlinear solver failures during online shape fitting, leading to transient error spikes. Planned refinements involve adopting robust real-time filters (e.g., Unscented Kalman Filters) and incorporating hysteresis models for improved compensation of tendon friction and preload variation. The next phase involves validating the MR-guided ablation workflow in vivo (e.g., with swine models and embedded RF tracking coils).

A plausible implication is that integrating Cosserat-rod kinematic models and analytical Jacobian derivations, together with continuous feedback from MR imaging, positions MRI-actuated robotic catheters as high-bandwidth and highly accurate steering platforms for complex cardiac interventions. The demonstrated sub-millimeter closed-loop accuracy and mechanical accessibility in the MR environment align with the stringent requirements for clinical translation in MR-guided electrophysiology and interventional radiology (Wang et al., 2023, Hao et al., 28 Dec 2025).