Staff Writer

Ensuring that preventable surgical mishaps don't happen are a top priority for hospitals in their ongoing quest to improve patient outcomes and reduce costs. Now, researchers from Johns Hopkins University in Baltimore, Md., and Siemens AG, based in Munich, have developed an algorithm to accurately identify spinal vertebrae during surgery. The expectation is that the algorithm, known as Levelcheck, could reduce, or even virtually eliminate, errors caused by misidentification of the spinal vertebrae during surgery; these are relatively rare roughly one in 3,000 cases of spinal surgery.

The algorithm is currently in clinical testing as a surgical decision support tool, and could ultimately become a software add-on that could be added as update to surgical radiograph machines from device makers including Siemens.

"It's 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," said Steven Krosnick, director of the National Institute of Biomedical Imaging and Bioengineering (NIBIB) program in Image-Guided Interventions. The research was funded in part by a grant from the NIBIB, part of the NIH.


"Levelcheck is a good example of an idea that comes around when engineers are working alongside surgeons. Johns Hopkins University is one of the best places in the world for that – my lab is in a hospital," Jeffrey Siewerdsen, a professor of biomedical engineering and computer science at Johns Hopkins, told Medical Device Daily.

"Everyone hears about wrong side surgery the knee or hand on the wrong side of the body," he continued. "It was only in conversation with a spine surgeon that we discovered this was even a thing. At some degree of frequency, surgeons operate on the wrong level of the spine. If you can see the sacrum, you can count but anything in between is very difficult."

The researchers first started working on Levelcheck in 2012 under a research partnership between Siemens and Johns Hopkins. The algorithm works by pairing a previously taken 3-D image, such as a CT or a MRI, and patches it to the 2-D radiograph taken at the beginning of surgery.

Within 20 to 40 seconds, the algorithm compares the images, matching the previously labeled CT or MR to positions and landmarks, and projecting labels onto the X-ray image. The process takes less than a minute, scarcely interrupting the surgical work flow. That's a feature that's crucial for surgeons.

"Most of the time it is just confirming something that you would have gotten right anyway," Siewerdsen said. "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."


Siewerdsen expects the algorithm could be used routinely as decision support to verify the usual results achieved by visually counting and finding the vertebrae. That would provide a higher level of certainty of accuracy, as well as set the surgical team's mind at ease.

If it's demonstrated successfully enough, Levelcheck could even theoretically obviate the need for manual counting in the operating room of spinal vertebrae. But any initial FDA clearance would likely be as an adjunct to the current standard procedure.

The algorithm is currently in clinical testing in the surgical suite as a check to the initial assessment. The initial evaluation is compared to the algorithm's results pre-surgery to determine the extent to which they agree, understand the source of any variance and introduce any corrections as necessary.

In a prior retrospective study, published in October in the journal Spine, the researchers analyzed 398 cases comparing the Levelcheck results with the assessments of a panel of three spine surgeons. The surgeons found that the assessment was helpful in 42 percent of these cases, improved their confidence in another 30 percent and did not ever diminish performance. Levelcheck was also found to be 100 percent accurate in vertebral labelling in this study.


The algorithm was particularly useful when images lacked obvious anatomical landmarks, vertebrae are obscured by other anatomy such as the shoulders, there is poor radiographic image quality, there are anatomical variations or abnormalities, and for level counting across long spine segments.

Obese patients, in particular, often have poor radiographic image quality. The Levelcheck process enables surgeons to clearly identify their vertebrae during surgery despite that impediment.

"When you have an obese patient, you have a much lower translation of image quality. It tends to be mostly black, you can't see the features you really want. That can challenge a human who is counting their way along," said Siewerdsen.

The researchers also are conducting a laboratory study evaluating how the algorithm is best used in practice either as decision support or as the primary means of evaluation. Siewerdsen noted that the researchers have already received several inquiries from institutions across the U.S. that are interested in using Levelcheck at their hospitals.

"The algorithm was developed in collaboration with Siemens. So, it could be translated to a Siemens product, since they make X-ray machines. It could be broader than that as well. There are other companies who make X-ray systems for the operating room, we are free to translate that as broadly as we wish," he said.

"We are interested in seeing that inter-clinical translation in a broad form. In principle, it could be just a software add-on for existing machines," added Siewerdsen.

The researchers are already working on developing other applications for the 3-D to 2-D registration technology that underlies Levelcheck. It could be used to visualize the placement of medical devices during surgery as well as enable the collection of quantitative surgical data.

"If you put a screw in, for example, it needs to be positioned very accurately, particularly if you are near major vessels or nerves," said Siewerdsen. "If I have a 3-D model of that screw, which I have, I can lay it very precisely with the 2-D to 3-D image registration and see is that screw where you intended it to be?"

No Comments