Surgical Robot
3D illustration of surgical robot on white background [3alexd/ E+/ Getty Images]

Robots can be trained with artificial intelligence to autonomously assist during laparoscopic surgery, researchers report, with the system successfully tested in large animal model.

The study, described in Science Robotics, is the next step towards the development of automated surgical copilots that could assist doctors during operations.

Using an open-source, AI simulator, the surgical robots were trained to autonomously perform multi-step, procedures such as gauze picking and clipping blood vessels during laparoscopic, robot-assisted operations.

The learning-based solution was independent of the robotic platform and worked with both the da Vinci Research Kit and the Sentire Surgical System.

“Surgical robotics holds immense potential for advancement by building upon embodied AI,” maintained researcher Yonghau Long, a PhD student at The Chinese University in Hong Kong, in a video presentation of the findings.

Nonetheless, she added: “Ethical guidelines and regulatory frameworks should be established given that [the] ‘surgical copilot’ is emerging at a pace faster than anticipated.”

Surgical robots are increasingly used in minimally invasive procedures and autonomy is envisaged to improve efficiency under these growing demands, said Long.

As part of this, AI-powered approaches offer the potential of generalized automation across various surgical paths.

To achieve these goals, the researchers created a framework called VPPV, which encompasses visual parsing, a perceptual regressor, policy learning, and a visual servoing controller, in which the robot’s motion is guided by visual cues.

They tested this on seven, robotic skill-training “game-based” tasks that align with the Da Vinci skills-based curriculum including: needle reach; needle pick; gauze retrieve; pick and place; peg transfer; match board; and needle regrasp.

The system showed a robust performance in this real-world situation and was stable when varying the positions and orientations of the object.

To examine its feasibly for real surgical applications, the team then conducted ex vivo experiments on animal tissue for five surgically assistive tasks under the Sentire system.

These included: endoscopic tracking; needle grasping; gauze picking; soft tissue retraction; and blood vessel clipping. The tests were conducted with various object types, instrument types and illuminations and demonstrated the generalized ability of the method.

Finally, the investigators tested the system in vivo using a live porcine model to see its performance in various tasks during laparoscopic, robot-assisted surgery.

“Three tasks in dynamic, in vivo, surgical environments are validated including gauze picking, tissue retraction and vessel clipping, which demonstrates the applicability of our method in clinical practice,” Long reported.

She suggested that future areas of application for the system could include intelligent skill training, automating long-horizon complex tasks, and collaboration between humans and intelligent robots.

Also of Interest