By Aine Cryts
Vol. 22 No. 3 P. 14
Radiomics may uncover the path to personalized cancer treatments.
Eighty-four percent of patients with lung cancer—the second-most common cancer among men and women—have non–small-cell lung cancer (NSCLC). The good news is death rates due to lung cancer, including NSCLC, have declined by 51% for men since 1990 and 26% for women since 2002, according to the American Society of Clinical Oncology (ASCO).
Two reasons fewer people are dying from lung cancer are that they never start smoking or they quit the habit. A third reason, according to ASCO, is medical advances in diagnosis and treatment.
Two therapies used to treat NSCLC are tyrosine kinase inhibitors and immune checkpoint inhibitors. The challenge? Choosing the right therapy isn’t an easy decision for oncologists. That’s because biomarkers can change during a therapy course, thus rendering a treatment ineffective.
That’s where some promising research at Moffitt Cancer Center in Tampa, Florida, may help. There, researchers are using AI to maximize the information they capture from PET/CT scans of patients. This work may help oncologists determine the appropriate treatment path for their patients.
Matthew Schabath, PhD, an associate member at Moffitt Cancer Center, points out that other tests, such as tissue or liquid biopsies, require additional testing and may take between 24 hours and a week to generate findings. That’s in contrast to radiomics, which relies on images that are already taken of patients during their care journey.
Radiomics and Lung Cancer Treatment
Anyone with a smartphone can easily understand how radiomics works, Schabath explains. He points to the ability of an iPhone or an Android device to assess images and then classify them into groupings. For example, using AI, these devices absorb information about personal images—namely, by the specific family, friends, and pets in them—and organize them in a folder for us.
At a basic level, that’s how radiomics can assess images to determine the appropriate treatment regimen for patients. Specifically, Schabath’s research project used PET/CT imaging with an 18F-FDG radiotracer. Working together, the combination can detect areas of abnormal glucose metabolism and help characterize tumors, he says.
According to Schabath, this type of imaging is widely used in the staging of patients with NSCLC. Specifically, the radiotracer is known to be affected by the activation and inflammation of the epidermal growth factor receptor (EGFR), a mutation often found in patients with this type of cancer. The detection of EGFR can help oncologists determine the appropriate treatment path, Schabath says, since patients with an active EGFR mutation have a better response to tyrosine kinase inhibitor treatment.
A study by his research team, which was published in Nature Communications, detailed how an EGFR deep learning score was created using retrospective data from patients at Shanghai Pulmonary Hospital and Fourth Hospital of Hebei Medical University, both located in China. The researchers then validated the model using data from patients at Fourth Hospital of Harbin Medical University in China and Moffitt Cancer Center. To train the model, the researchers assessed the retrospective PET/CT scans of 429 patients. The model was validated among 187 patients, and 65 patients’ scans were used as an external test.
In an announcement, Robert Gilles, PhD, chair of the cancer physiology department at Moffitt Cancer Center and a coauthor of the study, said, “We found that the EGFR deep learning score was positively associated with longer progression-free survival in patients treated with tyrosine kinase inhibitors and negatively associated with clinical benefit and longer progression-free survival in patients being treated with immune checkpoint inhibitor immunotherapy.”
Gilles added that, while he’d like to perform additional studies, this approach could help inform different treatments as a clinical decision-support tool.
Schabath stresses that this is a study based on retrospective data. A prospective trial is the logical next step. Although securing funding for the trial will be challenging, he’s not unduly concerned. Using this technology in patient care is possible within the next five to eight years, but it could take as long as 10 years. “But 10 years is the very top end of this,” Schabath insists.
Mining Images for Cervical Cancer Treatment
Beth Erickson, MD, FACR, a professor in the department of radiation oncology at the Medical College of Wisconsin in Milwaukee, where she’s also chief of brachytherapy services, notes that radiation oncologists have long been dependent on imaging. Historically, she and her radiation oncologist colleagues have relied on imaging to understand the location of a patient’s cancer, thereby helping assess the area of the body that will be irradiated and influencing the success of the therapy; CTs, MRIs, and PET scans are very helpful with this, she adds.
But she’s equally enthusiastic about radiomics because calculations can be done on the images that are captured after a patient receives radiation. For example, AI can mine the images for information about whether the tumor is dead or alive, she explains. Radiomics can also be used to determine whether a patient is likely to be a long-term survivor or likely to have a recurrence. “It’s a whole other layer of digging into the imaging world that we, as radiation oncologists, can use,” Erickson says.
Radiation oncologists are vigilant for a scenario where the radiomic data say the tumor still isn’t dead after six months of radiation and that a higher dose is recommended. “[With that information,] we still have the time and opportunity to deliver [additional treatment] based on that data,” she says.
In current practice, it can be difficult to discern whether the tumor is responding, Erickson explains. Radiomics may give radiation oncologists a tool to understand whether they should intensify or decrease treatment, she adds.
According to the Centers for Disease Control and Prevention, women are far less likely to die as a result of cervical cancer—due to the use of regular Pap tests—than they were 40 years ago. Still, the American Cancer Society projects that approximately 14,480 new cases of invasive cervical cancer will be diagnosed and approximately 4,290 women will die from cervical cancer in 2021.
Erickson is currently running a retrospective study to determine whether radiomics can help inform patients’ treatment plans. The study includes 87 scans captured from 43 patients whose cervixes were imaged in the same MRI machine since 2009. She explains that women with cervical cancer receive five weeks of external radiation, after which they typically receive five brachytherapy treatments.
After each of the brachytherapy treatments, MRI scans of the patient’s cervix are taken. To obtain the images, the women are put under anesthesia and an applicator is inserted into their uterus against their cervix. The MRI images then guide the radiation oncologist’s treatment plan. All of the MRI images that have been captured of the study participants since 2009 are now being evaluated for radiomic signatures in a computer program created by one of Erickson’s colleagues at Medical College of Wisconsin.
“We’re looking for clues in the radiomic data to see what could be a good feature and what’s a bad feature. We’re looking at things like local control survival, and we’re trying to relate what’s not seen but what’s analyzed with artificial intelligence,” Erickson says. The goal is to find out whether any of those factors are prognostic, she adds.
Erickson emphasizes that she’s not using radiomics for patients who are under active treatment. Her research, which is still in an exploratory stage, is helping to guide what radiation oncologists should look for when radiomics is a part of regular practice.
One of the challenges with this work is the requirement for large patient datasets to validate the findings, Erickson says. In addition, all of the patients’ MRI scans must be captured on the same MRI scanner. That’s because the software programs that run on scanners will vary from machine to machine, she explains. Medical College of Wisconsin is a good location for this study because all of the women’s images have been captured on the same 3 T MRI machine.
“Our dataset should be quite pure because of that, but one of the things that you find is you need scans on patients done on the same scanner over time to really know if there’s any feature that’s more important in terms of their prognosis. That’s hard to come by,” Erickson says.
She doesn’t have a timeframe for when radiomics could be part of clinical practice for the treatment of women with cervical cancer, and there are currently no clinical trials scheduled to prospectively study its use among this patient population.
Erickson’s enthusiasm for the future of radiomics—and the impact it may have on patient care—is palpable. “We get kind of enthused about things that interest us, but it’s because we can personalize future patient care if we know more. It just helps us so much to understand. If we need to intensify treatment for a patient to cure them, I’d like to know that early on so I can do the best job of that and not harm them with excessive doses to their internal organs,” she says.
“The sooner we know in the course of someone’s treatment that they may be at higher risk and need more intensified treatment—whether that’s because they get started or as they’re getting treated—the better,” Erickson explains.
Schabath shares her enthusiasm for the promise of radiomics. He’s encouraged by the uptake of radiomics in the research community and among oncologists and physicians in other specialties.
“If you don’t have buy-in from clinicians, you’re just spinning your wheels. We’re seeing clinicians understand the importance of this field,” he says. “I feel fortunate to be part of an important area that I really think is going to improve the care of patients in the future.”
— Aine Cryts is a health care writer based in the Boston area.