An Image-Guided Robotic System for Transcranial Magnetic Stimulation: System Development and Experimental Evaluation

TL;DR

This work tackles the variability and geometry-driven inaccuracies in transcranial magnetic stimulation (TMS) coil placement by introducing an image-guided robotic system (RoTMS) that uses finely segmented brain meshes for standardized pose reporting. It combines a complete hardware/software pipeline—segmentation, registration, pre-operative planning with deterministic pose extraction from curved brain surfaces, and three scalp-projection heuristics—with sub-millimeter registration and hand-eye calibration, plus 2D/3D magnetic-field sensing for validation. The major contributions are three deterministic pose-planning heuristics, demonstrated improvements in actuation accuracy (positional errors halved; rotational accuracy up to two orders of magnitude) and reduced variability across trials, and higher, more stable induced magnetic fields. These results support more reproducible, geometry-aware TMS experiments and set the stage for standardized reporting of coil poses in research and clinical settings.

Abstract

Transcranial magnetic stimulation (TMS) is a noninvasive medical procedure that can modulate brain activity, and it is widely used in neuroscience and neurology research. Compared to manual operators, robots may improve the outcome of TMS due to their superior accuracy and repeatability. However, there has not been a widely accepted standard protocol for performing robotic TMS using fine-segmented brain images, resulting in arbitrary planned angles with respect to the true boundaries of the modulated cortex. Given that the recent study in TMS simulation suggests a noticeable difference in outcomes when using different anatomical details, cortical shape should play a more significant role in deciding the optimal TMS coil pose. In this work, we introduce an image-guided robotic system for TMS that focuses on (1) establishing standardized planning methods and heuristics to define a reference (true zero) for the coil poses and (2) solving the issue that the manual coil placement requires expert hand-eye coordination which often leading to low repeatability of the experiments. To validate the design of our robotic system, a phantom study and a preliminary human subject study were performed. Our results show that the robotic method can half the positional error and improve the rotational accuracy by up to two orders of magnitude. The accuracy is proven to be repeatable because the standard deviation of multiple trials is lowered by an order of magnitude. The improved actuation accuracy successfully translates to the TMS application, with a higher and more stable induced voltage in magnetic field sensors.

Figures (5)

• Figure 1: Neuro-navigation system and pre-operative planning strategies. Panels a. and b. illustrate the comparison between our proposed system and an existing commercial system. Our system uses a fine-segmented brain mesh model to maintain geometrical details. This information can be used to determine the approaching pose of the TMS coil so a reference (true zero) can be established. In contrast, the existing commercial system uses a radially sliced volume image to estimate the boundary of the brain. It lacks details, and the pose of the TMS coil cannot be automatically calculated based on the true geometry. Panel c. shows the strategies to automatically calculate a pose on a mesh surface (Section \ref{['sec:toolplansurface']}). Each of these strategies defines both a position, determined by a center point and an orientation, which is established in different ways. The 4-point constraint uses a center point along with three additional points, including one that determines the tail direction, to form a plane that defines the orientation. The 3-point constraint selects one of three points as the center position. Lastly, the 2-point constraint searches the triangle within the mesh model that contains the specified center point to determine the orientation. Using any of the above constraints, Panel d. shows three different TMS coil pose calculation methods (Section \ref{['sec:toolplanstrategies']}). Free Skin Pose uses the shape of the skin and gets a pose perpendicular to the scalp. It uses no cortical geometrical information except the human estimation of the target's location. Restricted Cortex Pose obtains an orientation tangential to the target cortex. Closest Skin Pose uses the target location in the cortex, projects it to the closest point on the skin, and obtains a tangential pose to the skin. Note these three strategies are not individually corresponding to the strategies in panel c. Each strategy in panel d. may use any pose calculation method in panel c. Panel e. shows examples using our proposed system.
• Figure 2: The architecture and the kinematics of the proposed image-guided robotic TMS system. The Main Control PC hosts the user interface and the robotic TMS (RoTMS) control packages. The user interface visualizes the medical images, segmented models, and real-time poses of the robot end-effector. The pre-operative plans and the robot command are also entered from the user interface. The RoTMS package manages the workflow and communications between the controller and the user interface. The Operation Scene contains a robotic arm, a controller, an optical tracker, and a human subject with a rigid body marker attached to the head. Arrows and reference frames illustrate the complete kinematic chain. From left to right, these reference frames are {R} robot base, {E} end-effector, {C} center of the TMS coil, {Cr} rigid body marker on the TMS coil, {O} optical tracker, {H} head of the subject, and {Hr} rigid body mark on the head. {$R\rightarrow E$} is obtained by robot sensors, {$O\rightarrow Cr$} and {$O\rightarrow Hr$} by the optical tracker, {$E\rightarrow Cr$}, {$Cr\rightarrow C$}, {$Hr\rightarrow H$} by registration and calibration (Section \ref{['sec:registration']}), and {$H\rightarrow b$} by pre-operative planning (Section \ref{['sec:toolplan']}). TMS and EMG setup shows the TMS stimulation controller and data recording devices and their connections. More details are provided in Section \ref{['sec:overview']}. This figure contains graphical elements licensed from BioRender.com.
• Figure 3: The design and the recording methods of the 2D/3D magnetic field sensors. Panel a. is the design of the 2D magnetic field sensor and a setup using a head phantom. The 2D sensor can measure one axis of the applied magnetic field, and the measurement is a scalar. In the phantom, the brain can be used as a localization tool so that the location of an attached 2D sensor is known when covered by the 3D-printed skin shell. An example of the oscilloscope measurement is shown between panels a. and b. Panel b. is the design of the 3D sensor that can measure all 3 axes of an applied magnetic field in space. The measurement results can be represented in vectors. Panel c. shows a physical setup using the proposed sensors to measure the magnetic field generated by a TMS coil actuated by our robotic system (robotic arm not shown). Rigid body markers are attached for the purpose of navigation.
• Figure 4: The results from the alignment tests in the human subject study. From left to right, S1-3 are the 3 subjects. The Pose Error panel shows the errors between the planned and measured poses in the 10 alignments for each method. The error transformation is from the planned pose to the measured pose, and all 3 axes are shown. Geometrically, the z-axis is the perpendicular direction of the skin surface, and the y-axis is in the direction of the tail of the TMS coil. The poses are represented in small vectors where the translation is the coordinate of the small cross, and the orientation is the tilt of the vectors. The size of the cluster represents repeatability. The Error Distribution panel shows the violin plots of each group of 10 alignments. The results in this panel are shown in Euclidean distance and the angle component of the angle-axis representation of a 3D rotation. The left sub-column shows the translational errors, and the right shows the rotational errors. The means and the standard deviations are provided in the table at the bottom of each row.
• Figure 5: The results of the coil-holding test of the phantom head and 3D sensor voltage measurement. Panel a. shows the sensor voltage measured at the same pose from sending 20 trains of the pulses over 5 minutes. Each sub-panel shows the results in one axis from three actuation methods, robotic, manual operation 1 and 2. The data of manual operator 2 contains missing traces in the secondary axes, and those results are skipped in the plot (gaps in the yellow line). Panel b. shows the mean and standard deviation, organized by axes and actuation methods. The numbers in the table are color-coded in the same colors as the lines of the different actuation methods in Panel a. Panel c. illustrates that a small sensor displacement causes a change of the received magnetic flux, where the more accurate method to have lower readings in the primary axis, and the less accurate one may have higher readings in the secondary axes. The illustration matches the results in Panel a. This figure contains graphical elements licensed from BioRender.com.