Post-surgical lost and found5/14/2015
Producing 2,800 cadaver images sounds like a macabre undertaking, but the cause is worthwhile. After all, you wouldn’t want anything accidentally left behind in your body after surgery, right? It’s thought to happen somewhere between 3,500 and 30,000 times a year, though it’s hard to get a definitive count.
This isn’t due to the carelessness of surgeons. They rarely lose anything – roughly once in about 3,000 surgeries that carry a risk for lost objects (think chest/abdominal/emergency). But with 20 to 34 million such surgeries per year, the cases add up. And the incidents cost $1.5 billion annually.
That sounds pricey, but systems designed to prevent these mistakes are even more expensive to implement, so hospitals haven’t done it. Instead, teams typically count the surgical tools before and after the procedure and make sure everything that went in comes back out. And we humans are imperfect counting machines.
While some hospitals, such as the Mayo Clinic, routinely x-ray to double check that the body cavity is clear of surgical tools, most only x-ray if the count is off or considered unreliable. And even when these x-rays happen, radiologists often can’t spot objects like surgical sponges - gauze pads that are used to sop up blood and other liquids or move innards out of the way for a less unobstructed work space.
Flexible metal ribbons embedded in the sponges help them to appear in x-rays at all, but the sponges are still disguised by conforming to body cavities. As many as half of them are missed in x-rays. And yet the sponges are particularly insidious, since organs such as intestines can grow into them and develop holes.
“As a resident, we saw these cases,” said Theodore Marentis, a former radiology resident at U-M. “We saw both the few missed objects and how easy it would be to miss them.”
Motivated to find a better method for preventing lost objects, he initiated a collaboration with engineers to develop a tag that shows up clearly in post-surgical x-rays but is also small enough to attach to sponges.
First, he needed to find an engineer who could create such a tag. As luck would have it, he already had a contact. “I knew Nikos socially, through the Greek society,” said Marentis, quickly clarifying, “Not the fraternities - actual Greek.”
“Nikos” is Nikolaos Chronis, an associate professor of mechanical engineering as well as biomedical engineering. He and Marentis designed a simple tag that could be attached to surgical sponges. Originally, it was comprised of three tiny balls of tungsten carbide, a material whose density lends itself to armor-piercing bullets. But more to the point, high density also makes it show up brightly in x-ray images.
Chronis set the balls, just under a millimeter in diameter, into an FDA-approved plastic to produce a tag that would show up in the x-ray image as a triangular arrangement of bright dots. Then, Marentis tested it with a radiology phantom, basically a density model of the human body that looks realistic in x-ray images. The dots appeared very clearly.
In December 2011, Marentis presented his results at the U-M hospital in the radiology Grand Rounds, the annual gathering at which department researchers share their work. That’s how he met Lubomir Hadjiiski, a research professor in radiology.
Hadjiiski believed that software he and Heang-Ping Chan, a professor of radiology and pioneer in computer aided diagnosis systems, had developed to detect breast cancer would also work to detect the tags. The software searches for tiny deposits of calcium in breast tissue, which appear as bright dots in x-ray images and can signal the presence of a tumor.
The tag could show up in different orientations, so it doesn’t look exactly the same in every x-ray image, but the range of possibilities is much smaller than the natural variation in the calcium deposits. This means the software could search for tags with higher accuracy than is possible for tumors.
Shortly after Hadjiiski discovered the project, Marentis and Chronis decided to switch to a three-dimensional tag. If the tag was side-on in the x-ray image, it was possible for two balls to overlap, making the tag much more difficult to spot. Chronis added a fourth ball, producing a pyramid shape.
“However the tag is oriented in 3D space, we can always visualize at least three spheres,” said Chronis. “It’s easy to identify with the eye or with the software.”
In the summer of 2012, Marentis tested the tags by placing them on and under a human cadaver and taking x-rays, so the tags in the images look almost as though they were inside. They needed a large bank of images in order to develop and test the software, so Marentis took about 2,800 of them.
Then, Hadjiiski and Chan redesigned the automated detection software so that it could find the tags in the x-ray images. The first stage of processing identifies as many potential tags as possible. The second sorts out the real tags from the false positives – sutures, for example.
That second stage uses an artificial intelligence strategy to decide whether suspected tags are real or not. It is trained on images in which tags and false positives have been identified. The algorithm figures out how to tell them apart after analyzing many training images, about 500 in this study.
The trained program can run in two modes: one for the surgeon and one for the radiologist. For the surgeon, it can locate 80 percent of the tags and only sees a tag where there is none about 0.3 percent of the time. This means that if the software turns up a tag, the surgeon knows to stay in the operating room until the radiologist can respond.
For radiologists, the software can turn up 90 percent of the tags, but about one in five images contains a false positive. To the expert eye, the false positives are easily dismissed.
Radiologists would see the computer’s identifications only after studying the x-ray images themselves. This is the typical procedure to avoid prejudicing the radiologist into missing a tag not seen by the computer.
But that reasoning shouldn’t stop a surgeon from benefiting from the software’s immediate response, Hadjiiski pointed out. “It seems there is an opportunity for the computer to be the first reader in the operating room,” he said. In that case, the surgeon could begin removing a lost sponge without waiting around for the radiologist.
To test whether the software works well with radiologists, the team had radiologist volunteers try to spot the tags with and without help from the software. Want to know how they did? Watch this space. For about a year. Yeah, I know, medical journals are slow, but that’s the price of rigorous review.
The tag must jump a lot of hurdles before it will see an operating room, but Chronis and Marentis are trying to make it happen. They founded the company Kalyspo to manufacture the tags and are currently seeking collaboration with a surgical sponge manufacturer. Once they have some tagged sponges, they’ll need a hospital that is willing to use them.
If the tag ever achieves wide usage, the software’s accuracy should improve over time as more images containing the tags are fed into the training set. And it may even be expanded. Surgical sponges account for roughly two thirds of the objects left behind in patients, and needles make up another 25 percent. Hadjiiski and Chan are working on teaching the software to see them as well.
Chronis is also an associate professor of macromolecular science and engineering.
About Michigan Engineering: The University of Michigan College of Engineering is one of the top engineering schools in the country. Eight academic departments are ranked in the nation's top 10 -- some twice for different programs. Its research budget is one of the largest of any public university. Its faculty and students are making a difference at the frontiers of fields as diverse as nanotechnology, sustainability, healthcare, national security and robotics. They are involved in spacecraft missions across the solar system, and have developed partnerships with automotive industry leaders to transform transportation. Its entrepreneurial culture encourages faculty and students alike to move their innovations beyond the laboratory and into the real world to benefit society. Its alumni base of more than 75,000 spans the globe.