Learning-Based Modeling of a Magnetically Steerable Soft Suction Device for Endoscopic Endonasal Interventions

TL;DR

The paper tackles the challenge of real-time, precise shape control for magnetically actuated soft robots in endoscopic endonasal neurosurgery. It introduces a learning-based framework that directly maps magnetic-field inputs to a compact Bezier-curve representation of device shape, leveraging embedded FBG shape sensing to provide ground-truth-like deformation data. Random Forests outperform Neural Networks, achieving sub-millimeter accuracy in Bezier control-point prediction and shape reconstruction, and revealing interpretable feature importance tied to magnetic fields, frequency, and distance. The approach offers a practical, interpretable, and fast pathway for real-time closed-loop control of magnetically steered soft surgical tools and can generalize to other magnetically actuated continuum robots, with future work focusing on confined environments, hybrid physics-informed learning, and ex vivo validation.

Abstract

This paper introduces a learning-based modeling framework for a magnetically steerable soft suction device designed for endoscopic endonasal brain tumor resection. The device is miniaturized (4 mm outer diameter, 2 mm inner diameter, 40 mm length), 3D printed using biocompatible SIL 30 material, and integrates embedded Fiber Bragg Grating (FBG) sensors for real-time shape feedback. Shape reconstruction is represented using four Bezier control points, providing a compact representation of deformation. A data-driven model was trained on 5,097 experimental samples to learn the mapping from magnetic field parameters (magnitude: 0-14 mT, frequency: 0.2-1.0 Hz, vertical tip distances: 90-100 mm) to Bezier control points defining the robot's 3D shape. Both Neural Network (NN) and Random Forest (RF) architectures were compared. The RF model outperformed the NN, achieving a mean RMSE of 0.087 mm in control point prediction and 0.064 mm in shape reconstruction error. Feature importance analysis revealed that magnetic field components predominantly influence distal control points, while frequency and distance affect the base configuration. Unlike prior studies applying general machine learning to soft robotic data, this framework introduces a new paradigm linking magnetic actuation inputs directly to geometric Bezier control points, creating an interpretable, low-dimensional deformation representation. This integration of magnetic field characterization, embedded FBG sensing, and Bezier-based learning provides a unified strategy extensible to other magnetically actuated continuum robots. By enabling sub-millimeter shape prediction and real-time inference, this work advances intelligent control of magnetically actuated soft robotic tools in minimally invasive neurosurgery.

Figures (12)

• Figure 1: Magnetically steerable soft suction device for endoscopic endonasal brain tumor resection. (Top) Conceptual illustration showing how the device is inserted through the nasal cavity and sphenoid sinus using a standard suction endoscope to reach deep-seated skull-base regions. (Bottom) Cross-sectional schematic illustrating the device’s mechanical design and dimensions. The soft body (gray) encloses a suction lumen (2mm) and a channel routing the multi-core Fiber Bragg Grating (FBG) sensor at the neutral axis for accurate curvature sensing. The outer diameter is 4mm, and the active flexible length is 40mm. A hollow cylindrical permanent magnet (red/blue) at the distal tip enables magnetic steering under externally applied rotating magnetic fields. Additional geometric and material specifications are provided in Table \ref{['tab:device_specs']}.
• Figure 2: FBG-reconstructed shapes (black dashed lines) and corresponding Bezier curve fits (colored solid lines) for four representative frames during 1.0 Hz magnetic field rotation with the soft robot tip positioned 90 mm above the coil table.
• Figure 3: Experimental setup for magnetic actuation and shape sensing of soft robot. The system includes an FBG-embedded soft robot mounted on a linear actuator, positioned above electromagnetic coils at a distance of $d_{tip}$. A power supply and stepper driver control the linear actuation, while real-time strain data is collected via the FBG interrogator and visualized on a workstation. Close-ups on the right show the FBG clamping, suction tube, 3D printed fixtures, and the soft robot with a tip magnet for steering.
• Figure 4: Time-varying magnetic field components ($b_x$, $b_y$) for the 0.2 Hz trials corresponding to the trajectories in Fig. \ref{['fig:shape3d']}.
• Figure 5: 3D shape reconstruction (0.2 Hz). Each curve represents a centerline shape reconstructed from FBG data at different magnetic field strengths.