Less Margin for Error
By Beth W. Orenstein
Vol. 20 No. 9 P. 10
Optical imaging methods show promise for delineating tumor margins.
During surgical resections, surgeons typically rely on their eyes and fingertips to probe where a tumor begins and ends. After removing what they believe and hope is the entire tumor, they take samples of surrounding tissue to send to pathology. Then, they and the patient wait—as long as several weeks—for the pathology report to determine whether the margins are clear. If the margins are not clear, the surgeon may have to go back and remove more tumor and surrounding tissue. Someday soon, however, imaging devices may change this paradigm.
“There’s a tremendous amount of activity in the area of using imaging devices to determine tumor margins,” says Bruce Tromberg, PhD, director of the National Institute of Biomedical Imaging and Bioengineering at the National Institutes of Health. Tromberg and his colleagues agree that the technology is promising. “This is an area that’s quite logical and will continue to grow and expand,” he says.
“There’s no question that within the next five years, we’re going to have an imaging technology that’s going to be in later-stage clinical trials, and it’s going to have a high degree of sensitivity to detect disease at the margin during surgeries so that we can assure we get clear margins with one operation,” says Hank Schmidt, MD, PhD, FACS, an associate professor of surgery at the Icahn School of Medicine at Mount Sinai in New York.
There’s already one device on the market—MarginProbe by Dune Medical—that has been cleared by the FDA for use in identifying positive margins during breast cancer surgery in real time. The technology uses radiofrequency spectroscopy to probe tissue and identify cancer cells that may be remaining on the surface of the breast tissue that has been removed. Imaging devices that will help surgeons determine margins also are in development for other solid tumors including those of the colon, prostate, and brain as well as head and neck cancers. Schmidt is investigating another device—Otis Wide Field optical coherence tomography (OCT) by Perimeter Medical Imaging—for use in lumpectomies, while his Mount Sinai colleague, Brett Miles, DDS, MD, FACS, a professor of otolaryngology at Icahn and cochief of the division of head and neck oncology for the Mount Sinai Health System, is looking at the same device for head and neck cancer surgeries.
Two Workflow Approaches
Tromberg says researchers around the world are looking at two workflow approaches using imaging devices to determine margins. One is imaging directly in the surgical field. “This approach could be done with a camera-type system or handheld probe, and both technologies, with and without the addition of contrast agents, are widely under development,” Tromberg says.
There is precedent for this approach, he notes. When looking for sentinel lymph nodes to dissect, surgeons typically inject a blue dye and/or radioactive tracer and rely on visual guidance or a handheld radiation-sensitive probe to find the nodes. “The contrast agent helps guide the surgeon to the node, and that’s where you cut,” Tromberg says. “But the presence or absence of cancer is still determined by traditional pathology methods.”
The second workflow approach is to remove the specimen and then image it in the operating room to assess the margins. The Otis Wide Field OCT uses this approach. “We excise the tumor as a single specimen,” Schmidt says. “Then we hand it immediately to the technologist who can image it with the OCT device. This particular instrument scans at very high speed so we can start looking at the images almost right away. It’s very similar to scrolling through any type of axial data set such as CT or MRI. We have the ability to examine the entire surface of a three-dimensional piece of tissue.”
The OCT device scans the tumor on six sides. “Then we correlate, with a fair amount of specificity, exactly where on that particular side there may be disease close to the margin indicating additional tissue still needs to be removed,” Schmidt says.
In the 18 months that his project has been underway, Schmidt and his colleagues have used the device on about 100 patients undergoing lumpectomies for both invasive carcinoma (IC) and ductal carcinoma in situ (DCIS), an early-stage cancer. They are not yet using the device to make interoperative decisions in real time.
“We’re collecting data with OCT and comparing it to what we’re seeing in the pathology reports when they come back,” Schmidt says. In the first 50 patients or so, “we saw good correlation—nearly 97% of cases had similar results.”
At some point, OCT will be used to make decisions on additional surgery in real time, Schmidt says. At that point, he says, the hope is that answers about the need for further surgery will be the result of automated reads. But, he adds, “We have discussed making the data available immediately in our PACS system so that radiologists could go online, look at the data as soon as they’re captured, and communicate to the surgeon where, if any, additional tissue should be excised.” Radiologists may also contribute to the data used to determine the automated results, Schmidt notes.
Otis Wide Field OCT
Miles is running a similar study using the Otis Wide Field OCT device for head and neck cancers. Like Schmidt, Miles says, “At the current time, we’re not changing our procedures. We’re imaging the specimens, gathering the data, and then performing routine frozen sections like we always would. Eventually, if we gather enough data and the results are what we expect, there’s the potential for the system to give a readout and tell the surgeon whether the margin is clear or not. We’d have to prove that it’s at least as good as the pathologist’s report. But that’s what we’re working toward.”
It helps, Miles adds, that the image resolution is extremely high. “We are really able to see the microarchitecture of the tissues with this system because of its ultrahigh resolution,” he says. “The resolution has the ability to detect very subtle differences even a human eye can’t see.”
The device should work particularly well for head and neck tumors because “what’s unique about tumors taken from the tongue and tonsils is that once you remove all that tissue, all you should see is normal muscle or fat on the deep surface,” Miles says. It’s a simple yes/no question, and normal tissue looks benign on these images.
“So, I’m suspecting that this system could perform even better for what we want to do than for breast tumors, which are somewhat more complex,” Miles says. Another advantage to this system, he says, is that it has depth. “A lot of imaging platforms are surface only.”
Miles has been looking for optical imaging techniques he can apply to head and neck cancers because the traditional method of sending tissue to pathology isn’t always the best approach. “A 4 cm X 4 cm surface of frozen section isn’t ideal, and by the time you freeze these sections and look at them in pathology, a lot of error can occur,” he says. “Little areas can be missed, and that’s why I’ve been interested in optical imaging research.”
Miles says head and neck surgeons could still rely on frozen sections but use optical imaging to target sections, making the samples more accurate. “Currently, if frozen margins are positive, we remove more tissue where we think tumor cells might remain,” he says. “If I had a system that enabled me to image the entire tumor surface, we would have a higher level of accuracy of the location of the remaining tumor and know whether we need to take out more tissue in that specific region.”
Researchers in the Netherlands are studying hyperspectral imaging for margin assessment after breast-conserving surgery to improve surgical outcomes. Hyperspectral imaging, which originated from remote sensing and has been used for various applications by NASA, is an optical imaging technique that can measure the entire resection margin within a limited amount of time, without tissue contact or the need for exogenous contrast agents, says Esther Kho, a PhD student at The Netherlands Cancer Institute in Amsterdam and lead author of a study published in Clinical Cancer Research in June 2019. Hyperspectral imaging measures diffusely reflect light: The tissue is illuminated with normal halogen light sources, and the light undergoes multiple scattering and absorption events in the tissue. With a hyperspectral camera, the light that is scattered back to the surface of the tissue is captured. Thereby, an “optical fingerprint” of the tissue is obtained that reflects the composition and morphology of the tissue, which can be used for tissue analysis, Kho explains.
Hyperspectral cameras differ from normal cameras in that they capture 256 images of the tissue at different wavelengths, whereas normal cameras, eg, cell phone cameras, capture images in only three major spectral bands. Therefore, subtle differences at specific wavelengths can be seen. In addition, hyperspectral cameras capture light at the near-infrared wavelength range.
“Therefore, we can measure tissue characteristics that are invisible to the human eye—specifically, the amount of water, fat, and collagen in the tissue,” Kho says. “Since tumor and healthy tissue are optically different in this wavelength range, we are able to differentiate them.”
In their study, the researchers imaged breast tissue slices, which were obtained after gross sectioning of the resected breast specimens. “We obtained a data set that was highly correlated with histopathologic information,” Kho says. “We showed that, on this data set, we could discriminate tumor from healthy tissue with high accuracy.”
Their ultimate goal, however, was to image the resection margins of breast specimens. So, instead of imaging gross-sectioned tissue slices, they imaged the surface of the specimens before gross sectioning. In this study, the researchers imaged six unsectioned specimens to evaluate hyperspectral imaging of these specimens in the current clinical workflow.
“As mentioned in the manuscript, the correlation of hyperspectral analysis with histopathology on these specimens was difficult, due to the limited available histopathologic information,” Kho says. “Therefore, we were not able to verify the whole measured surface, and we could not get a classification performance of hyperspectral imaging for resection margin assessment.”
The researchers are currently performing a study in which this problem is solved: By analyzing hyperspectral images immediately after imaging, locations that are suspected to be tumor positive and negative can be marked and retrieved after histopathologic processing. In this way, a direct correlation with histopathology can be obtained, and the classification performance of hyperspectral imaging for resection margin assessment can be evaluated. “Currently, we are analyzing these measurements,” Kho says.
The study group includes patients who have had primary breast surgery. “Thereby, we obtained data on both IC and its potential precursor, DCIS,” Kho says. “We showed that for both cancerous tissue types, the classification accuracy was high. Therefore, the technique could be used for resection margin assessment during all breast cancer surgeries in which the surgeon would like to know if IC or/and DCIS is present at the surface.”
Kho believes this technique could definitively be used for other types of cancer surgeries as well. “In order to use this technique during other cancer surgeries, tumor tissue needs to be optically different from the surrounding healthy tissue,” she says. “Our group also has published promising results in detecting cancerous tissue in colon and head and neck specimens.”
Role for Radiologists
Tromberg finds it interesting that the radiologic community is not necessarily following these developments. He says that’s largely because the enabling technical advances are coming mostly from the optics/photonics and bioengineering communities. Still, Tromberg sees a role for radiologists in developing optical imaging tools, especially as they expand not only from the operating room but also to the bedside. His advice to radiologists: Get in on it while the tools are in their early stages.
Some of the tools, Tromberg predicts, will require new ways to develop and interpret mechanisms of tissue contrast for optimum decision making. “And, when you have contrast, the images aren’t a simple black-and-white type of thing. There are plenty of gray areas in between. That’s where computational methods like artificial intelligence come into play, and the radiology community can help facilitate the decision making that is necessary to build an AI library.”
The radiology community is well versed in contrast media and developing radiomics that could apply, he says. “The radiology community has put a lot of work and effort into developing quantitative imaging biomarkers, so this is a kind of natural progression.”
— Beth W. Orenstein of Northampton, Pennsylvania, is a freelance medical writer and regular contributor to Radiology Today.