AI robot operates on gallbladder without human assistance

Surgical robots can also perform complex tasks autonomously

Learning AI surgeon: A newly developed surgical robot can surgically remove a gallbladder without the assistance of human surgeons. The AI ​​system learned this through videos of operations and targeted feedback during practice surgeries. This allowed the robot surgeon to develop its skills similar to that of a young doctor. The innovative surgical robot can now operate independently and reacts confidently even to unexpected complications during surgery.

The surgical robot removing a gallbladder. © Juo-Tung Chen/Johns Hopkins University

Research in the field of medical robotics has made great progress in recent years. Operations with assistive robots are already performed thousands of times every day. Many of these surgical assistants can also navigate autonomously within the body. For example, in 2022, a robot operated independently on a living animal for the first time – a laparoscopic operation on a pig.

However, this Smart Tissue Autonomous Robot (STAR), developed by a team led by Axel Krieger at Johns Hopkins University in Baltimore, still required special markings in the tissue to be operated on, worked only in a strictly controlled environment, and followed a rigid, predefined surgical plan. The robot was mechanically precise and worked carefully, but could not accommodate changes to the planned procedure.

Robot Learns Gallbladder Removal

Researchers led by Krieger and his colleague Ji Woong "Brian" Kim have now developed a new robot: the Surgical Robot Transformer-Hierarchy (SRT-H). It is based on artificial intelligence and can therefore learn interactively, similar to ChatGPT. This allows it to flexibly adapt to the situation during surgical procedures, unlike its predecessor: It recognizes individual anatomical features, makes spontaneous decisions, and corrects itself if things don't go as expected.

This AI-assisted surgical robot was initially trained using videos of operations. Surgeons from Johns Hopkins University performed the operations on 34 pig cadavers specifically for this purpose, recording them via cameras on their wrists and subsequently adding captions describing the procedure. From this extensive video footage, the robot learned to perform three basic, short surgical tasks: guiding a needle, lifting body tissue, and suturing.

It then learned to surgically remove a gallbladder – a complex sequence of 17 individual tasks lasting several minutes – using voice-guided experiments on lifelike models. The robot learned to identify and precisely grasp specific blood vessels and the gallbladder duct, strategically place clips, and sever pieces of tissue with scissors. Through feedback such as "move your left arm a little to the left," it improved and was eventually able to perform the operation with 100% accuracy, the team reports.

The Surgical Robot Transformer-Hierarchy (SRT-H) performs surgical procedures, adapts in real time to individual anatomical features, makes spontaneous decisions, and corrects itself when things don't go as expected.© Juo-Tung Chen/Johns Hopkins University

Robot Surgeon Adapts to Unexpected Procedures

In total, Kim and his colleagues tested the robot under eight different starting positions and surgical conditions. The models – consisting of the gallbladder and the surrounding tissue – were not anatomically uniform and each appeared different due to the addition of blood-like dyes. This simulates emergencies that can occur during surgery.

But even in these situations, the robot performed flawlessly, adapting its movements to the changing conditions and correcting its own errors, as the researchers report. Although the robot took an average of around five minutes to complete the operation, longer than a human surgeon, its results were comparable to those of an experienced surgeon. "This shows that robots can fundamentally perform complex surgical procedures independently," says Krieger.

Will AI robots soon be in the operating room?

The team concludes that an AI robot can be trained to become a skilled surgeon in a similar way to a young intern: through demonstration and imitation. The new system thus combines the mechanical precision of a robot with the understanding and adaptability of a human, as the team explains.

"This advancement takes us from robots that can perform specific surgical tasks to robots that truly understand surgical procedures," says Krieger. "This is a critical distinction that brings us significantly closer to clinically viable autonomous surgical systems that can function in the chaotic, unpredictable reality of actual patient care."

According to the researchers, their prototype demonstrates that autonomous robots could soon operate in hospital operating rooms. "Our work shows that AI models can be made reliable enough to enable surgical autonomy—something that once seemed far away, but is now demonstrably practical," says Kim.

Training for Additional Surgery Types

The team is now training and testing the SRT-H system for additional types of operations to expand its capabilities. The long-term goal is to perform fully autonomous surgery on real people in the future.

However, one hurdle in the development of such surgical robots is the training material, writes Michael Yip of the University of California, San Diego, in an accompanying article. Due to data protection concerns, there are currently hardly any videos available of real surgeries on human patients with which the AI ​​systems could be trained. Manually labeled videos and models or simulations, as in this case, could potentially provide a solution.

(Science Robotics, doi: 10.1126/scirobotics.adt5254 / doi: 10.1126/scirobotics.adt0684)

Sources: American Association for the Advancement of Science (AAAS), Johns Hopkins University