Imagine a physician attempting to reach a cancerous nodule deep within a patient's lung – a target the size of a pea, hidden behind a maze of critical blood vessels and airways that shift with every breath. Straying one millimeter off course could puncture a major artery, and falling short could mean missing the cancer entirely, allowing it to spread untreated.
This is the high-stakes reality physicians face in thousands of procedures daily, where accuracy is critical and the task is complicated by anatomical obstacles that are non-penetrable or sensitive. Can artificial intelligence (AI) and robots help address these challenges and improve patient outcomes?
"A new era of "AI guidance" is dawning in medicine," says Ron Alterovitz , Lawrence Grossberg Distinguished Professor in the Department of Computer Science. "Robots with advanced AI can assist physicians and automate certain tasks, enabling unprecedented levels of accuracy and making complex procedures safer and more effective."
A recent article in Science Robotics formalizes the concept of AI guidance for medical needle-based procedures and outlines the degrees of AI guidance for each component. The authors define four components of AI guidance: perceiving anatomy, planning instrument motions, perceiving instrument state, and performing instrument motions during a procedure. Robots with AI guidance can achieve greater accuracy and precision in needle guidance than human physicians, while also enabling the use of state-of-the-art needle designs that can curve to steer through the body.
The paper, "Medical Needles in the Hands of AI: Advancing Towards Autonomous Robotic Navigation," was written by Alterovitz, as well as Janine Hoelscher of Clemson University and Alan Kuntz of the University of Utah. Hoelscher and Kuntz both previously completed doctorates in computer science at UNC with Alterovitz as their adviser.
The era of AI guidance
For decades, physicians have relied on image guidance–using x-ray images, computed tomography (CT) scans, and magnetic resonance (MR) images to visualize a patient's anatomy and plan needle paths prior to a procedure. This advance, dating back to the discovery of x-rays in the late 1800s, enabled safer access to points within the body.
Recent advances in AI are now enabling a leap forward. AI can automatically analyze images, identify targets and obstacles, compute safe trajectories, and even steer robotic needles around sensitive tissues to sites deep in the body autonomously. One such example, shown by a team of researchers at the University of North Carolina at Chapel Hill, Vanderbilt University, and the University of Utah, demonstrated a medical robot that can autonomously steer a needle to clinically-relevant targets in a lung with high accuracy in living tissue, demonstrating better performance than physicians using only traditional tools. The authors detail this transformative shift from image guidance to AI guidance, with AI assisting in perceiving patient anatomy, tracking progress through the procedure, planning instrument motions, and even performing those motions.
"For decades, image guidance has helped physicians to better plan and execute medical procedures," Alterovitz said. "AI guidance is going even further toward making procedures safer and less invasive."
The article formalizes the concept of AI guidance, which leverages AI to enhance physician performance and create building blocks for higher levels of robot autonomy. The article defines the four components of AI guidance as:
perceiving anatomy
planning instrument motions
perceiving instrument state
performing instrument motions during a procedure
Each of these four components can provide its own degree of AI guidance:
Eyes-on/Hands-on -- where the physician performs the task with AI providing assistance
Eyes-on/Hands-off -- where the AI performs the task, while the physician monitors the AI and is prepared to intervene in unusual situations
Eyes-off/Hands-off -- where the AI performs the task and the physician only steps in when requested by the AI
Full AI Guidance -- where the AI fully performs the task
Finally, the article classifies current systems into these categories and discusses research challenges to enable higher degrees of AI guidance. In particular, the researchers highlight critical barriers to widespread clinical adoption, such as the need to guarantee safety, operation within the regulatory environment, the development of physician-AI interfaces that are intuitive throughout each degree of AI guidance, and the ability to integrate the necessary technology into all aspects of a clinical workflow.
While acknowledging the many challenges still to overcome, Alterovitz expressed excitement over the future of AI guidance in medical procedures.
"Breakthroughs in AI and robotics will continue to enable increasing degrees of AI guidance and robotic automation for medical procedures," Alterovitz said. "AI and robotics can provide physicians with new tools to make challenging procedures safer and more effective."
