It’s ‘trust, but verify’ for new AI spine surgery system

On a recent morning, Dr. Joseph Osorio arrived in the operating room ready to sink six surgical screws into his patient’s spine, and he did not seem remotely nervous that their placement and size had been recommended by artificial intelligence software.

Osorio was the first neurosurgeon on the West Coast to begin using Medtronic’s new “Stealth AXiS” surgical robotic system, conducting a spinal fusion procedure to treat degenerative scoliosis at Jacobs Medical Center in La Jolla by anchoring two small custom-shaped metal rods across three vertebrae in his patient’s lower spine.

The process started with a CT scan, identifying the segment of spine that needed reinforcement. A program analyzed the resulting three-dimensional image, using an AI model trained on information from previous successful surgeries, not just where screws should go, but also the best path for their insertion.

Board-certified with thousands of such surgeries in his past, having completed a fellowship at Columbia University after a surgical residency at UC San Francisco, Osorio is well-qualified to make these calls with zero help from technology. So, why bother using an algorithm to plan these crucial, but routine details?

The utility, he said, is similar to what many are now experiencing when they use AI writing tools. The software can quickly get a person to the neighborhood of what they intend to say.

“You might say, ‘write me a paragraph on this,’ and it’s going to cut down your time, but you might still need to change some words, add a comma, tweak a sentence … that’s essentially what the AI is doing here,” Osorio said.

In this particular case, the AI system’s recommendations for screw length and diameter seemed on point, allowing the army of surgical technicians assisting with the procedure to pull the proper supplies ahead of time. The suggested locations, though, did require minor adjustment.

“It was slight, very slight, I’d say probably, like one or two millimeter adjustments,” Osorio said.

And the AI auto planning feature, he added, is even more useful in situations where a patient’s health insurance company will not pay for a pre-surgical CT scan, meaning that the guiding image must be taken after the patient is already sedated in an operating room on the day of their surgery.

Once a digital surgical plan is created and approved by a qualified surgeon, a surgical robot can use a system of cameras and electromagnetic sensors, registered against each patient’s anatomy with an initial X-ray, to move its arm to each screw location, placing a drilling guide at the exact angle needed to put each anchor in the correct spot. Here, too, AI is at work comparing previously recorded X-rays with real-time sensor data to compensate for any patient movements that may occur.

It is an evolution of Medtronic’s previous “Mazor” robotic spike system, which had already achieved levels of anatomy navigation using sensors and cameras that have reduced the need for X-ray images during surgery. And other medical device companies have launched similar systems, building in AI functions as the entire industry begins to see such augmentation as table stakes to play in a market that has always been as competitive as a high-stakes table in a Las Vegas casino.

Patients may wonder whether this push toward AI guidance is a good thing. After all, this is a technology that has made headlines for its ability to “hallucinate” convincing, but incorrect details.

ECRI, an independent non-profit organization that works to improve patient safety and cost effectiveness in health care, has been watching these systems develop.

In an email, Scott Lucas, ECRI’s vice president of devices, therapeutics and technology, said that the organization does not comment on any individual case or procedure, but has found that AI-enabled systems do have their merits.

“We can say that AI seems to be particularly helpful when it is used to support imaging, planning, navigation and precision in technically demanding procedures such as spine surgery,” Lucas said. “These tools may help surgeons in multiple ways, including tailoring procedures to a patient’s anatomy and improving consistency in implant placement and alignment.”

That said, the executive makes it clear that there is no argument for blind loyalty. In these early days, he argues, AI assistance should be less involved with surgeries, and there must be a clear path that allows surgeons to verify the work that their algorithmic assistants perform.

“Used well, AI may strengthen surgical safety; used without appropriate governance, human oversight, training and monitoring, it could introduce new risks, including overreliance, workflow disruption, planning errors or automation bias,” Lucas said.

Such bias, he added, occurs when a surgeon “fails to recognize when the technology is wrong.”

Osorio said that he believes the checks and balances built into the new system he is now using weekly do give him solid checkpoints to make sure that the machine is not hallucinating. While screw placement calculations will automatically calculate for straightforward placements, those with particularly complicated circumstances, such as anatomy that significantly deviates from the norm, will not proceed.

“If things aren’t lining up perfectly in the image, or they’re getting some feedback, it will just refuse to place a screw in that corridor,” Osorio said. “So, it’s only making recommendations in locations that meet the highest standards, and it still requires the surgeon to validate every level.”

AI is now also involved in the calculations used to move the robotic arm to the correct locations for screw insertions and also to make real-time corrections for any patient movement, Medtronic confirmed by email.

Here too, Osorio said, there are ways to verify that the robot’s calculations are pointing at the correct vertebrae, even though this type of minimally invasive surgery does not expose the target bone before screw insertion.

Surgeons use a bony projection at the back of each vertebra called the spinous process to check the robot’s accuracy, laying a special navigation ring over the landmark to verify that what is showing on the computer’s calculated location screen matches the robot’s arm position.

“A very common statement is ‘trust, but verify’,” Osorio said.

While robotic spine surgery is the latest to begin the process of AI integration, other systems have already made similar moves in knee and hip replacements, urologic procedures, and in some aspects of general surgery.