Data-driven adaptive needle insertion assist for transperineal prostate interventions

Abstract

Objective.Clinical outcomes of transperineal prostate interventions, such as biopsy, thermal ablations, and brachytherapy, depend on accurate needle placement for effectiveness. However, the accurate placement of a long needle, typically 150-200 mm in length, is challenging due to needle deviation induced by needle-tissue interaction. While several approaches for needle trajectory correction have been studied, many of them do not translate well to practical applications due to the use of specialized needles not yet approved for clinical use, or to relying on needle-tissue models that need to be tailored to individual patients.Approach.In this paper, we present a robot-assisted collaborative needle insertion method that only requires an actuated passive needle guide and a conventional needle. The method is designed to assist a physician inserting a needle manually through a needle guide. If the needle is deviated from the intended path, actuators shifts the needle radially in order to steer the needle trajectory and compensate for needle deviation adaptively. The needle guide is controlled by a new data-driven algorithm which does not requirea prioriinformation about needle or tissue properties. The method was evaluated in experiments with bothin vitroandex vivophantoms.Main results.The experiments inex vivotissue reported a mean final placement error of 0.36 mm with a reduction of 96.25% of placement error when compared to insertions without the use of assistive correction.Significance.Presented results show that the proposed closed-loop formulation can be successfully used to correct needle deflection during collaborative manual insertion with potential to be easily translated into clinical application.

Keywords: data-driven model; medical robotics; needle insertion assist; transperineal prostate intervention.

Figure 1.

In-bore MRI-guided prostate biopsy as an example of transperineal prostate intervention. (a) The physician is inserting a biopsy needle through the patient’s perineum on the MRI table using a grid template. (b) Axial and reformatted coronal slices of T1-weighted MRI acquired after the needle placement (upper) and 3D models of the prostate and the planned and resultant needle path reconstructed from the image (lower). Interactions between the needle and layers of tissue cause it to deviate from the planned straight-line, resulting in the large targeting error.

Figure 2.

A simple 2DOF device can be used to position a needle guide that simulates the function of a transperineal prostate grid template. The robot is placed in front of the patient’s perineum and moves the guide in the $\mathrm{X}-\mathrm{Z}$ plane to reach the entry point planned from MRI images. In the pictured example (Song et al. 2013), the robot comprises a lead-screw-and-nut mechanism that transmits vertical and horizontal motion (blue arrows) to two crossbars. At the crosspoint of the bars, there is a passive needle guide for the manual insertion of the needle on the $\mathrm{Y}$-axis direction.

Figure 3.

In the assisted needle insertion, the operator pushes the needle manually in the $\mathrm{Y}$-axis direction. During insertion, however, the needle deflects from the straight-line planned path. Hence, the passive needle guide is automatically displaced in $\mathrm{X}$- and $\mathrm{Z}$- directions, generating and input $\mathrm{\Delta }{\mathbf{\Phi }}_k$ (left). As a result of the guide displacement, the needle tip is affected by a change on its current pose, described by the output $\mathrm{\Delta }{\mathbf{\Psi }}_k$, and realigned to the straight-line planned path (right).

Figure 4.

System architecture. (a) Overall schematic of the system. (b) Software components present in the control computer. In blue, the ROS2 package implemented in this work.

Figure 5.

(a) Phantoms used to validate the proposed trajectory correction by radial displacement. The phantoms were placed in a plastic container with the fabric layer surface facing an opening through which needle insertions were performed. (b) 2 DOF needle guide robot used for experiments.

Figure 6.

Experimental setup used to evaluate our proposed method. The stage extremity is zoomed in to show the needle inserted through the passive needle guide and entering into the tissue phantom.

Figure 7.

Estimation results using an arbitrary ${\stackrel{\mathbf{ˆ}}{\mathbf{J}}}_0$. (a) Random radial displacement applied at the needle guide. (b) Correspondent needle tip positions in robot coordinates. In blue, our model estimation. In red, actual values measured by the EM tracker.

Figure 8.

Data-driven model estimate errors for prediction of the needle tip position when using different initial Jacobians. (a) Experimentally obtained ${\stackrel{\mathbf{ˆ}}{\mathbf{J}}}_0$. (b) Arbitrarily defined ${\stackrel{\mathbf{ˆ}}{\mathbf{J}}}_0$

Figure 9.

Trajectory errors for needle insertions into phantom A. In blue, 5 insertions in open-loop. In red, 5 insertions with closed-loop MPC control.

Figure 10.

Example of a stepwise 100mm length needle insertion into ex-vivo phantom with (red) and without (blue) MPC closed-loop control. (a) Displacements applied at the needle guide. (b) Correspondent needle tip positions in both vertical and horizontal directions.