Respiratory Motion Compensation and Haptic Feedback for X-ray-Guided Teleoperated Robotic Needle Insertion

TL;DR

This paper tackles respiratory motion as a key source of error in X-ray-guided percutaneous liver procedures by integrating real-time motion estimation from an external EM surrogate with robotic needle steering for motion compensation, paired with proximity-based haptic feedback for remote insertion. The approach comprises a motion-estimation stage trained against fluoroscopic ground truth and a teleoperation stage that guides insertion while providing tactile cues. Validation on a liver phantom shows motion-estimation MAEs below 3 mm in the main motion axes and quantified 3D insertion errors, highlighting reduced radiation exposure due to remote operation. The work indicates practical potential to improve targeting accuracy and safety in respiratory-affected percutaneous interventions, while outlining areas for future improvement such as tissue interaction effects and in vivo validation.

Abstract

Respiratory motion limits the accuracy and precision of abdominal percutaneous procedures. In this paper, respiratory motion is compensated robotically using motion estimation models. Additionally, a teleoperated insertion is performed using proximity-based haptic feedback to guide physicians during insertion, enabling a radiation-free remote insertion for the end-user. The study has been validated using a robotic liver phantom, and five insertions were performed. The resulting motion estimation errors were below 3 mm for all directions of motion, and the overall resulting 3D insertion errors were 2.60, 7.75, and 2.86 mm for the superior-inferior, lateral, and anterior-posterior directions of motion, respectively. The proposed approach is expected to minimize the chances of inaccurate treatment or diagnosis due to respiratory-induced motion and reduce radiation exposure.

Figures (3)

• Figure 1: (a) The respiratory motion compensation step uses an electromagnetic tracker attached to the liver phantom to create a motion model. The model is trained using as ground truth the motion of the target inside the liver phantom extracted from fluoroscopic images. The robotic arm with a needle at its end-effector compensates for the respiratory motion using the output of the motion model. The robot performs needle steering to maintain the alignment between the target and the needle. (b) In the remote needle insertion step, the end-user remotely controls the robotic arm by operating the haptic device with force feedback ($F_{\text{user}}$), guiding the physicians to reach the desired displacement ($\textit{x}_{\text{user}}$).
• Figure 2: (a) A transformation matrix from the base frame ($\Psi_B$) to the end-effector ($\Psi_{EE}$) realigns the robotic arm with the frame of the moving target ($\Psi_T$). The arm moves based on the estimations made using the electromagnetic (EM) system as a surrogate. The estimations relate the real-time position of the sensor in the EM frame ($\Psi_{EM}$) to the target frame ($\Psi_T$). (b) The source side contains the haptic device controlled by the user to move the robotic arm. The user moves the handle in the $y_o$ direction relative to the initial location of frame ($\Psi_O$). (c) The replica side contains the liver phantom with a spherical lead target located inside the phantom. The EM tracker system is attached to the phantom and used as a surrogate signal, and the C-arm scan is used to train and validate our methods. The robotic arm with a needle at the end-effector enables respiratory motion compensation and remote insertion steps.
• Figure 3: (a) The liver phantom simulates different types of breathing patterns during the training phase. The target location (ground truth) is extracted in the two directions of motion Anterior-Posterior and Superior-Inferior which are represented by the solid lines in blue and gray, respectively. The motion models are trained to estimate the ground truth (the dotted lines). (b) The liver phantom simulates regular breathing and the location of the target is extracted for validation (solid lines). The motion models estimate the current target position (dotted lines), and the robotic manipulator compensates for the respiratory motion by steering the needle attached to the end-effector (dashed lines). (c) The insertion is performed remotely using the haptic device. The forces felt by the user increase as the distance between the tip of the needle and the target decreases.