Algorithm provides an extra level of assurance during spine surgery


Researchers funded by the US National Institute of Biomedical Imaging and Bioengineering (NIBIB) have developed a way to automatically label images of individual vertebrae during spinal surgery, preventing mistakes and saving surgeons both time and stress in the operating room. New work recently published by the team demonstrates the accuracy, feasibility, and advantages of having the technology in the operating room.

Because of obesity, low bone density or anatomical abnormalities, the twenty-four separated vertebrae, sacrum, and tailbone can be difficult to distinguish and number. Both time and money are spent ensuring operations are performed at the right locations. Still, in about one out of 3,000 procedures, a mistake is made and the wrong level is operated on, causing unnecessary damage and requiring an additional, corrective surgery.

To reduce both the chance of error and the burden on the surgeons, a team from Johns Hopkins University, Baltimore, USA, and Siemens Healthcare in Germany have developed an algorithm dubbed “LevelCheck” to help identify and label vertebrae in real-time during surgery.

“It is more than just avoiding those one in 3,000 cases. It actually provides assurance to the surgeon, so it means they can be more confident. It just makes for a better procedure,” says Steven Krosnick, director of the NIBIB program in Image-Guided Interventions. It fits into the surgical workflow and doesn’t require much additional time during surgery, he says.

“When you look at a radiograph of the spine, it is difficult to say with 100% certainty what you’re looking at. The potential for human error is one that an algorithm could help to avoid,” says Jeffrey Siewerdsen, professor of biomedical engineering and computer science at Johns Hopkins University and senior author of the paper. “Surgeons spend a lot of time, energy, and stress to get it right, and we wanted to provide some decision support for that.”

The algorithm can label levels on radiographs

The researchers first described the algorithm in 2012. It takes advantage of two types of routinely taken images: a computerised tomography (CT) scan taken prior to surgery and a radiograph taken at the beginning of surgery. The CT can be used to accurately define and label the vertebrae. When surgery is set to begin, LevelCheck compares the current radiograph to the previously labelled CT, matches positions and landmarks, and projects the labels onto the radiograph. Using high-speed computing, the algorithm makes its comparison and provides labels in the span of 20 to 40 seconds.

In the new work, published in Spine, the researchers examined the usefulness of the LevelCheck algorithm by applying it to nearly 400 images previously taken from spinal surgery patients. Three spine surgeons evaluated both the algorithm’s accuracy and how useful they thought such a tool would be during surgery. LevelCheck labelled the vertebrae correctly in every case and the surgeons judged it to be helpful in 42% of the cases and to improve confidence in 31%. The algorithm was particularly advantageous when anatomical landmarks usually used to count spine segments, such as the sacrum or twelfth rib, were missing, obscured, or abnormal; when spine segments were not easily distinguishable; and when the image quality of the radiographs was poor. As for the additional time spent waiting for the labels to appear, the surgeons said they would be willing to wait up to a minute for the extra assurance.

Siewerdsen compares LevelCheck to GPS in cars; you rely on it when you are driving somewhere new, but you might also use it as a check or confirmation even when going places you have been before. “Most of the time it is just confirming something that you would have gotten right anyway,” he says. “But decision support can help you reach that decision a bit faster, with a bit more certainty. And every once in a while, it could even help prevent an error.”

The team has also designed a version that can be used when only preoperative MRIs, rather than CT scans, are available. Siewerdsen also sees potential for the technology to track and guide devices during surgery and to provide easier ways to collect quantitative data about surgeries.

The current work is retrospective. Still, Siewerdsen thinks LevelCheck could one day be commonplace in the operating room, installed into imaging systems so labels—and the precision and confidence they provide—are just a button push away.