Spinal fusion surgery requires highly accurate implantation of pedicle screw implants, which must be conducted in critical proximity to vital structures with a limited view of the anatomy. Robotic surgery systems have been proposed to improve placement accuracy. Despite remarkable advances, current robotic systems still lack advanced mechanisms for continuous updating of surgical plans during procedures, which hinders attaining higher levels of robotic autonomy. These systems adhere to conventional rigid registration concepts, relying on the alignment of preoperative planning to the intraoperative anatomy. In this paper, we propose a safe deep reinforcement learning (DRL) planning approach (SafeRPlan) for robotic spine surgery that leverages intraoperative observation for continuous path planning of pedicle screw placement. The main contributions of our method are (1) the capability to ensure safe actions by introducing an uncertainty-aware distance-based safety filter; (2) the ability to compensate for incomplete intraoperative anatomical information, by encoding a-priori knowledge of anatomical structures with neural networks pre-trained on pre-operative images; and (3) the capability to generalize over unseen observation noise thanks to the novel domain randomization techniques. Planning quality was assessed by quantitative comparison with the baseline approaches, gold standard (GS) and qualitative evaluation by expert surgeons. In experiments with human model datasets, our approach was capable of achieving over 5% higher safety rates compared to baseline approaches, even under realistic observation noise. To the best of our knowledge, SafeRPlan is the first safety-aware DRL planning approach specifically designed for robotic spine surgery.

Ultrasound (US) has been increasingly used as medical imaging technology across various clinical diagnostic and therapeutic scenarios thanks to its availability and non-radiative nature. While 3D US probes are becoming available, most systems are still using 2D probes. For 3D US reconstruction based on 2D probes, US image calibration forms an essential step. Through calibration, one can find the transformation matrix between a coordinate frame attached to an optical marker or the robot’s end effector towards the coordinate frame of the US probe. Current US calibration methods usually require hereto lengthy free hand gestures as well as some manual interventions, which hampers the use and integration with advanced robotic systems. This paper introduces a reliable automatic calibration framework that is also fast. Demonstrated on a KUKA lightweight robot and 2D US probe, the full calibration procedure was completed in 224.8 seconds with a 1.29 mm mean 3D localization error. Within this procedure, camera-to-robot calibration was accomplished within only 47 seconds and reached a 0.17 mm mean error. Validation of the US image calibration was done through 3D printed model, leading to a mean deviation of 1.05 mm from the respective CAD models.

Ultrasound (US) calibration is an important step in robotic 3D US reconstruction. It determines the spatial transformation from the tracking sensor frame to the US image frame. Up to now, many free-hand calibration methods have been developed. Some of them were applied in robotic systems. However, there is no comparative analysis for those employed methods in the robotic US system. Moreover, the required robotic scanning trajectory was normally unclear in previous work. Therefore, in this paper, we provided a comparative analysis of three popular US image calibration methods, namely cross-wire phantom, sphere phantom and Z phantom. Meanwhile, we attempted to provide concrete examples of using three phantoms to establish automatic robotic calibration frameworks. Demonstrated on a KUKA lightweight robot attached with a 2D US probe, different aspects were assessed to show the calibration performance and 3D US reconstruction accuracy. With the proposed robotic calibration frameworks, the total time of entire calibration procedure was shortened within 3 minutes for all three phantoms. The reconstruction results showed that the RMSE were less than 1.6 mm. The proposed automatic robotic US image calibration frameworks together with their quantitative analyses provide a foundation for further development and optimization. It could potentially facilitate the development of robotic US system.

Minimally invasive pedicle screw placement (MIPSP) is a widely used treatment for spine diseases. When coupled with intraoperative navigation modalities, robots may help improve surgical outcomes and reduce complications. With such a system, the application of pedicle screws has been expanded from needle insertion to the spine surgery. This paper investigates the possibility and feasibility of robot-assisted MIPSP based on ultrasound (US) guidance. The proposed system is non-radiative and fiducial-free, using purely image information to close the registration loop. Then the system automatically positions the drill tip to a planned screw trajectory and executes the drilling operation. Experiments were conducted on both ex-vivo lamb and human cadaver spines. An entry point accuracy of 2.39±1.41 mm, and orientation accuracy of 2.82±1.85∘ was found for 24 drilled trajectories on three lamb spines. On the ex-vivo human spine, the position error averaged 3.08±2.43 mm at the entry point and 4.05±2.62 mm at the stop point across 16 drilling instances. Moreover, a 87.5% success rate was reported by using Gertzbein-Robbins grade. The experimental results demonstrate the potential for offering a radiation-free alternative. Although restricted to cadaver trials, this work encourages further exploration of this technology to assist surgeons in maximizing performance in clinical practice.

In the research field of robotic spine surgery, there is a big upcoming momentum for surgeon-like autonomous behaviour and surgical accuracy in robotics which goes beyond the standard engineering notions such as geometric precision. The objective of this review is to present an overview of the state of the art in non-visual, non-radiative spine sensing for the enhancement of surgical techniques in robotic automation. It provides a vantage point that facilitates experimentation and guides new research projects to what has not been investigated or integrated in surgical robotics. Studies were identified, selected and processed according to the PRISMA guidelines. Relevant study characteristics that were searched for include the sensor type and measured feature, the surgical action, the tested sample, the method for data analysis and the system’s accuracy of state identification. The 6DOF f/t sensor, the microphone and the electromyography probe were the most commonly used sensors in each category, respectively. The performance of the electromyography probe is unsatisfactory in terms of preventing nerve damage as it can only signal after the nerve is disturbed. Feature thresholding and artificial neural networks were the most common decision algorithms for state identification. The fusion of different sensor data in the decision algorithm improved the accuracy of state identification.