Robot equipped with artificial intelligence achieves 100% success in complex surgical operations

Researchers are developing an advanced robotic system to improve the precision and dexterity of endoscopic submucosal dissection, a complex surgical procedure. Yuancheng Shao from Tongji University, Yao Zhang from the University of Leuven, Jia Gu and others. DESectBot is a new two-segment continuous robot designed to overcome the limitations of existing single-segment tools. Their work details the implementation of deep learning controllers, specifically gated recurrent units (GRUs), to effectively address the challenges of simultaneously managing robot position and orientation and controlling connected sequential segments. Demonstrating superior performance against conventional control methods in trajectory tracking and complex peg movement tasks, and validated through ex vivo ESD procedures, this study represents a significant advance toward more reliable and user-friendly robotic assistance in minimally invasive surgery.

Dual-segment sequential robotics and deep learning to enhance endoscopic submucosal dissection improves accuracy and control.

Researchers have announced a new dual-segment continuous robot called DESectBot designed to significantly improve the accuracy and effectiveness of endoscopic submucosal dissection (ESD). This minimally invasive procedure is essential for the treatment of early gastrointestinal cancers, requires considerable technical skill, and is often hampered by the limitations of existing single-segment robotic tools.
DESectBot addresses these shortcomings with a separate structure and integrated surgical forceps, giving the tip six degrees of freedom and improving lesion targeting. This development is in response to the growing global burden of gastrointestinal cancer, which accounted for 26% of all cancer diagnoses and 35% of cancer deaths worldwide in 2018.

This study details the implementation of a deep learning controller, specifically a gated repeat unit (GRU), to simultaneously control the position and orientation of a robot tip. These GRU controllers effectively manage complex nonlinear coupling between consecutive segments, a key challenge in such robotic systems.

Benchmarks against established control methods, Jacobian-based inverse kinematics, model predictive control, feedforward neural networks, and long short-term memory networks demonstrated the superior performance of GRU. For nested rectangle and Lissajous trajectory tracking tasks, GRU consistently outperformed all alternatives, achieving position/orientation root mean square error (RMSE) of 1.11mm/4.62° and 0.81mm/2.59°, respectively.

Further validation included orientation control at a fixed position, and the GRU achieved an average RMSE of 0.14 mm and 0.72°. The peg transfer task demonstrated the reliability of GRU, achieving a 100% success rate (120 successes out of 120 trials) with an average transfer time of 11.8 seconds, significantly outperforming the novice-controlled system.

Ex vivo demonstrations confirmed the ability of DESectBot to grasp, lift, and resect tissue, demonstrating sufficient rigidity to divide thick gastric mucosa and providing adequate surgical working space for large lesions. These results highlight the potential of GRU-based control to increase accuracy, reliability, and ease of use in ESD surgical training and ultimately improve patient outcomes.

Dual-segment continuous robot gated repeat unit control benchmark demonstrates improved performance in challenging environments

DESectBot is a novel dual-segment continuous robot with integrated surgical forceps developed to improve dexterity during endoscopic submucosal dissection (ESD). The robot features a separation structure that allows six degrees of freedom (DoF) at the tip, improving lesion targeting. Deep learning controllers, especially gated repeat units (GRUs), have been proposed to simultaneously control the position and orientation of the tip and effectively manage the nonlinear coupling between consecutive segments.

The performance of the GRU controller was rigorously benchmarked against Jacobi-based inverse kinematics, model predictive control, feedforward neural networks (FNN), and long short-term memory (LSTM) networks. To evaluate the accuracy of the GRU controller, nested quadrilateral and Lissajous trajectory tracking tasks were employed.

GRU achieved the lowest root mean square error (RMSE) in both tasks, recording 1.11mm/4.62° and 0.81mm/2.59°, respectively, demonstrating excellent tracking accuracy. Further evaluation includes directional control at a fixed position using four target poses. Here, GRU achieved an average RMSE of 0.14 mm and 0.72°, consistently outperforming all alternative control methods.

Next, a peg transfer task was performed to assess the reliability and speed of the GRU. GRU achieved a 100% success rate, with 120 successful transfers in 120 trials, and an average transfer time of 11.8 seconds, significantly outperforming the novice-controlled system. An ex-vivo ESD demonstration was performed to verify the surgical capabilities of the robot.

DESectBot successfully grasped, elevated, and resected tissue, confirming sufficient rigidity to divide thick gastric mucosa and providing adequate surgical working space for large lesions. These results demonstrate that GRU-based control significantly improves accuracy, reliability, and ease of use in ESD surgical training and potential clinical applications.

Precise robot control facilitates precise trajectory tracking and surgical manipulation, improving surgical outcomes.

Researchers developed DESectBot, a dual-segment continuous robot with integrated surgical forceps and six degrees of freedom to improve lesion targeting during endoscopic submucosal dissection (ESD). The GRU controller achieved the lowest root mean square error (RMSE) of position/orientation in nested square and Lissajous trajectory tracking tasks, recording values ​​of 1.11mm/4.62° and 0.81mm/2.59°, respectively.

For directional control at a fixed position across four target poses, GRU achieved an average RMSE of 0.14 mm and 0.72°, demonstrating superior performance compared to all alternative control methods tested. In the peg transfer task, the GRU-controlled DESectBot achieved a 100% success rate, with 120 out of 120 successful attempts and an average transfer time of 11.8 seconds.

This standard deviation of transfer times significantly exceeded the system controlled by novice users. The in vitro ESD demonstration confirmed the robot’s ability to grasp, lift, and excise tissue with the scalpel completing the cut, demonstrating sufficient rigidity to divide thick gastric mucosa and adequate surgical working space for large lesions.

These results demonstrate that GRU-based control significantly improves the accuracy and reliability of ESD surgical training scenarios. The DESectBot’s dual-segment design provides increased agility compared to single-segment robotic tools and facilitates access to complex anatomical areas such as the esophageal-stomach junction.

The robot’s performance in trajectory tracking and orientation control highlights the effectiveness of the GRU controller in managing nonlinear coupling between consecutive segments. The 100% success rate of the peg transfer task and the success of the ex vivo ESD procedure validate the potential for clinical application of this robotic system.

Gated iteration unit improves accuracy of continuous robot trajectory tracking by learning complex dynamics

A new dual-segment continuous robot named DESectBot was developed to improve precision and dexterity in endoscopic submucosal dissection (ESD). The robot features a separation structure and integrated surgical forceps that provide six degrees of freedom at the tip to enhance lesion targeting.

A deep learning controller based on gated recurrent units (GRUs) was designed to simultaneously manage the robot’s position and orientation and effectively deal with the nonlinear coupling inherent in continuous segment control. Benchmarked against Jacobi-based inverse kinematics, model predictive control, feedforward neural networks, and long short-term memory networks, the GRU controller consistently demonstrated superior performance in trajectory tracking and orientation control tasks.

Specifically, GRU produced the lowest root mean square error (RMSE) for both nested quadrangle and Lissajous trajectory tracking, and outperformed all alternatives in maintaining accurate orientation at a fixed location. The peg locomotion task yielded a 100% success rate, and ex vivo demonstrations in porcine gastric mucosa confirmed the robot’s ability to grasp, lift, and excise tissue with sufficient stiffness within a clinically relevant time frame.

The authors acknowledge the limitations of the control update rate, which is limited by the equipment used for ground truth measurements during the experiment. Future research will focus on increasing the control frequency through faster attitude sensing techniques and pipeline optimization. Further developments will also include the integration of tactile and shape sensing for improved feedback, real-time self-calibration, and standardized comparative studies with professional surgeons, alongside studies on workspace adjustment and collision avoidance with endoscopes. These advances aim to establish the DESectBot as a valuable tool for standardized ESD skills training and potentially improve surgical outcomes.