Toward a generalizable framework for robot-aided percutaneous surgery

Toward a generalizable framework for robot-aided percutaneous surgery Rosaura Morfino∗, Clemente Lauretti∗, Francesco Cocco∗, Francesco Prata†, Rocco Papalia†, and Loredana Zollo∗ ∗Research Unit of Advanced Robotics and Human-Centred Technologies, Campus Bio-medico University of Rome †Research Unit of Urology, Campus Bio-medico University of Rome [email protected] Abstract—Robot-aided percutaneous surgery faces significant challenges due to organ deformation between preoperative and intraoperative phases. This work presents an integrated framework including: deformable registration algorithm for aligning pre/intraoperative 3D models; real-time motion planner for needle path replanning; advanced teleoperated control strategy for needle insertion. The registration algorithm achieved an RMSE of 1.7±0.5mm with real-time suitability on clinical datasets. Validation of the planning and control modules on a phantom model with non-expert users showed improved needle orientation and positioning accuracy (maximum deviation: 0.006±0.003 rad, target error: 0.8±0.9mm). Clinical experts confirmed system feasibility with moderate cognitive workload. Index Terms—Percutaneous surgery, Surgical robotics, Image registration, Intraoperative planning, Teleoperated control I. INTRODUCTION Minimally Invasive Percutaneous Surgery (MIPS) enables access to deep anatomical structures through needle puncture, offering reduced patient trauma and improved outcomes compared to open approaches. However, intraoperative anatomical changes such as organ shifts, deformations, and positioning variations commonly compromise pre-planned needle paths. This limitation is particularly evident in procedures like percutaneous nephrolithotomy, where kidney displacement between supine preoperative imaging and prone intraoperative positioning significantly impacts the access route [1]. To address these challenges, several registration algorithms have been proposed to achieve accurate preoperative-tointraoperative alignment. Among these, geometric registration methods offer the optimal trade-off between accuracy and computational efficiency, enabling real-time alignment of preoperative 3D models with intraoperative data [2]. Once alignment is achieved, intraoperative replanning becomes essential to adapt needle trajectories based on updated anatomical information while maintaining patient safety. Teleoperated robotic systems represent a promising solution, as they preserve direct surgeon control while ensuring safe workspace constraints, avoiding the risks of autonomous systems in unmodeled scenarios and the radiation exposure inherent to cooperative approaches. However, existing teleoperated robotic systems for percutaneous procedures focus on needle insertion without consid- *This work was supported in part by the UCBM University Strategic Projects 2023 with the Proof of Concept (PoC) project BONE - Cooperative Robotic System for spinal surgery, and in part by the Piano Nazionale Ripresa e Resilienza (PNRR) - HEAL ITALIA Extended Partnership - SPOKE 2 Cascade Call - ”Intelligent Health” with the project BISTOURY - 3D-guided roBotIc Surgery based on advanced navigaTiOn systems and aUgmented viRtual realitY (CUP: J33C22002920006). ering preoperative-to-intraoperative alignment and real-time replanning capabilities [3]. This limitation represents a significant gap in current robot-aided percutaneous surgery frameworks. Therefore, this study introduces an integrated framework for robot-aided percutaneous surgery that combines pre-tointraoperative deformable registration with a novel motion planner for intraoperative trajectory replanning and a teleoperated control strategy for needle insertion, enabling surgeons to adapt to intraoperative anatomical changes while maintaining precise trajectory control. The framework is designed to operate independently of specific robotic platforms or procedure types, thereby supporting broad integration across surgical systems and percutaneous interventions. II. THE PROPOSED FRAMEWORK The proposed framework workflow is illustrated in Fig. 1a. Following 3D anatomical model reconstruction from medical imaging, a deformable registration algorithm aligns preoperative and intraoperative models to compensate for organ deformations and enable AR-based surgical guidance. Then, through teleoperation, a motion planner enables the surgeon to replan the trajectory of the needle connected to the robot based on the registered anatomy, followed by needle insertion. A. Deformable registration algorithm After 3D reconstruction from preoperative and intraoperative imaging using standard techniques from the literature [4], for example visual Simultaneous Localization and Mapping (SLAM) or deep learning–based approaches, deformable registration is performed through Coherent Point Drift combined with Optimized Volumetric Deformation (CPD+OpVD). Point clouds are extracted from both models, and a projectionbased partial selection identifies the portion of the preoperative model consistent with the intraoperative field of view. CPD estimates a non-rigid transformation through probabilistic correspondence matching, while OpVD propagates the deformation to the full preoperative model, producing a point cloud that accurately reflects intraoperative anatomy. B. Real-time motion planner Once intraoperative feedback is available through AR, the surgeon can replan the needle trajectory if required. During this phase, it is essential to preserve a safety margin from the patient’s skin and maintain the needle oriented toward the anatomical target. This is achieved through a novel motion planner based on a combination of a cylindrical motion planner and an orientation planner. The cylindrical planner maps haptic interface inputs along the xmand zmaxes 2025 I-RIM Conference October 17-19, Rome, Italy ISBN: 9788894580570 10.5281/zenodo.17629798 167 Fig. 1. a) Workflow of the proposed framework for robot-aided percutaneous surgery, b) advanced teleoperated control strategy for needle insertion. into an angular displacement θalong the cylindrical surface and an axial displacement ∆z(see Fig. 1a), constraining needle motion to a virtual cylinder. The orientation planner ensures that the needle axis remains aligned with the vector connecting the needle origin to the target at each instant through appropriate rotation matrices. During this phase, the haptic interface is governed by PD control, enforcing rigid interaction along the ymaxis, while the robot operates under conventional position control in the operational space. C. Teleoperated control architecture At this stage, the surgeon proceeds with needle insertion using a novel combined control strategy based on inverse dynamics control. The haptic interface is governed by a PD control law, which in this case enforces rigid interaction along the xmand zmaxes. The robot control architecture integrates aparallel position/velocity mode (S1) with a master–slave position error–based mode (S2), as illustrated in Fig. 1b. In S1, velocity control is applied along the insertion axis ys, while position control constrains xsand zsmotions to provide surgeon guidance; in S2, all three axes are controlled in position with maintained constraints on xsand zs. This combined approach was adopted since velocity control provides smoother motion, suitable for the initial insertion tract, whereas position control offers higher accuracy, making it preferable for target approach and final depth adjustment. The stabilizing action of the proposed control is expressed as: y=J† A(q)Ad−1KPAd˜ x−KDAd ˙ ˜x+KFFR (1) where J† Ais the right pseudo-inverse of the analytical Jacobian, Ad is the adjoint matrix mapping the pose error ˜ xto the end-effector frame, and KPand KDare the position control gains. The force term KFFRis present only in S1, with FRproportional to the velocity tracking error. The stabilizing action results in linear second-order error dynamics, which guarantee asymptotic stability under positive-definite gains and a nonsingular Jacobian. III. EXPERIMENTAL EVALUATION The registration algorithm was validated on two public liver datasets: i) DePoLL, comprising a porcine preoperative 3D model and 13 intraoperative deformed models, and ii) OpenCAS, including three simulated tests with preoperative and deformed intraoperative models and a phantom acquisition. An internal dataset from robot-assisted partial nephrectomy procedures was also used, with data from three patients including preoperative 3D kidney models and corresponding Fig. 2. a) Experimental setup, b) CPD+OpVD deformation example. intraoperative reconstructions. Performance, assessed in terms of execution time and RMSE reduction relative to rigid registration, achieved median values of 15 s[IQR: 6.5–38.5] and 72.95% [IQR: 63.4–82.5], respectively, confirming accuracy and real-time suitability for surgical workflows since the registration is performed once at the beginning of the operation (qualitative example shown in Fig. 2b). The planning and control modules were validated using the setup in Fig. 2a. Six non-expert users performed needle replanning and insertion in a percutaneous nephrolithotomy scenario on a gelatin-based anatomical phantom, under direct visual feedback. Evaluation metrics included orientation deviation from replanned trajectory, target positioning error (Euclidean distance between target and needle tip), and motion smoothness (mean/max velocity ratio). Results demonstrated maximum deviation of 0.006±0.003 rad, smoothness of 0.13±0.08, and target error of 0.8±0.9mm. Six clinical experts further assessed workload with NASA-TLX questionnaire, reporting a moderate average score of 58.2±12.8. IV. CONCLUSION This work presented an integrated framework that combines deformable registration, real-time replanning, and teleoperated needle insertion for robot-aided percutaneous surgery. Experimental validation demonstrated feasibility and robustness, while expert assessment confirmed its potential for clinical translation. Future work will address larger-scale studies and integration into surgical workflows. REFERENCES [1] A. Deshmukh et al., “Renal displacement with supine to prone positional change: Effect of sex and BMI,” Journal of Endourology, vol. 36, 2021. [2] Z. Han and Q. Dou, “A review on organ deformation modeling approaches for reliable surgical navigation using augmented reality,” arXiv. [3] M. Aggravi et al., “Haptic teleoperation of flexible needles combining 3D ultrasound guidance and needle tip force feedback,” IEEE Robotics and Automation Letters, 2021. [4] L. Zhou et al., “A Comprehensive Review of Vision-Based 3D Reconstruction Methods,” Sensors, vol. 24, no. 7, p. 2314, Apr. 2024. 168