Can Technology Help Predict Lung Cancer Prognosis?

In medicine and many other fields, technology is deepening our knowledge and advancing improvements. Such is certainly the case with lung cancer. Advanced technologies that leverage computer learning and new insights into how cancer works are leading to exciting developments that may one day make determining a precise diagnosis, prognosis and the best treatment for the individual as simple as scanning an image into a computer program or running a small sample of blood through a device.

Artificial Intelligence and Computational Image Analysis

Two recent studies have shown that advancements in artificial intelligence, or AI for short, and deep computer learning could someday lead to more precision in determining a lung cancer patient's fate.

Dr. Daniel L. Rubin, professor of biomedical data science, radiology and medicine (Biomedical Informatics Research) at Stanford University, led a study in the journal Nature in 2016 that found leveraging deep computer learning could improve pathologists' ability to determine a patient's prognosis. Typically a pathologist will diagnose lung cancer by looking at slides of a patient's cancer cells to find information about the type and stage of the disease. These assessments are based on the cancer cells' morphology -- their shape, the size of the cells, how they look when stained and other observable characteristics. But there are limitations to how well even the most experienced and well trained pathologist can see.

[See: 7 Things You Didn't Know About Lung Cancer.]

"The basic idea here is that pathology slides are looked at by pathologists to make diagnoses, and only a portion of the information that characterizes the disease is actually observable by the human eye. There's actually much more data in these pathology slides. This study was focusing on machine vision and AI techniques to leverage that information to get more clinically impactful information that can help with diagnosis and prognosis estimation to characterize the disease," Rubin explains.

The idea is if you can teach a computer what patterns to look for through the use of algorithms, it can compare the current slide against a much larger database of similar images and find a match that may provide detailed clues to the expected trajectory of the disease.

"The whole name of the game is we're recognizing that cancers are very heterogeneous," meaning that there's a lot of variation from one case to the next, Rubin says. "These labels we've been using to characterize cancer are too coarse. The whole basis of precision medicine is to more accurately characterize or differentiate these subtypes of diseases so better treatments can be given to patients as appropriate to the type of disease they actually have. These computer vision techniques that we and others are developing are tools to get at that."

Dr. Vamsidhar Velcheti, a thoracic oncologist at the Cleveland Clinic, recently earned a $3 million grant from the National Cancer Institute to complete a clinical trial using similar predictive analysis techniques to determine which patients are most likely to benefit from chemotherapy. The trial is based on a study Velcheti and his team conducted in collaboration with Case Western Reserve University in Cleveland. That research found that the technique, which focused on computerized image analysis of early-stage lung cancer tissue, could predict the likelihood of a recurrence in lung cancer patients.

Velcheti says that although patients with stage 1 or 2 lung cancer should be curable, and the intent of treatment -- typically surgery to remove the tumor and lymph nodes and chemotherapy -- is to cure the disease, nearly 50 percent of these early-stage lung cancer patients end up having a recurrence and dying of the disease. "This is very unlike breast cancer where the cure rates for stage 1 and 2 breast cancer are very high compared to what you see in lung cancer. The reason is lung cancer is much more aggressive and tends to go into the blood vessels."

Because the lungs are highly vascularized organs -- they need all that blood to send oxygen to the rest of the body -- even the presence of a single cancer cell left behind after chemotherapy can mean the cancer will come back. And all those veins and arteries mean any remaining cancer cells can travel more easily to distant locations where they might set up shop and create havoc. "If you have cancer of the lung, the chance of cancer spreading outside is much higher than if you have cancer in the breast or another organ. There's no other organ in the body that has as much blood supply as the lungs do," Velcheti says.

[See: What Not to Say to Someone With Lung Cancer.]

Currently, there's not a great way to determine which patients will have this problem with recurrence and which won't, which means most everyone gets the same blunt instrument of treatment with surgery and chemotherapy. The hope behind the advanced imaging clinical trial Velcheti is working on is to develop more precise ways of figuring out which patients are more likely to experience a recurrence so that clinicians can treat accordingly. "A lot of information that's embedded within the morphology of the cancer is lost because the human eye is limited by how much information it can actually detect," Velcheti says. "So using deep computational feature-extraction algorithms that we developed, you can develop signatures of the tumor morphology that could potentially inform us about how aggressive a patient's cancer is."

The analysis also looks at the tissue surrounding the tumor and how the immune system has responded to it for additional information about the biology of the cancer, how it's likely to grow and how the patient is likely to respond to chemotherapy.

All of this leveraging of computers to combine data sets and derive actionable information from myriad inputs doesn't mean that someday machines are going to replace humans in diagnosing cancer or determining how best to treat it. Far from it, Rubin says. "The goal is to improve the capability of the clinician and improve diagnostic accuracy by giving the patients as much information about their disease as possible and help guide treatment choices better than we can do," without the assistance of the vast data-crunching capabilities of a computer.

Liquid Biopsies

But it's not just pathology slides that could hold a key to predicting how lung cancer will change in individual patients over time. Cancer tumors shed cells into the bloodstream, and for the past several years, researchers have been studying the patterns associated with how these cells accumulate in the blood stream with a technique called liquid biopsy.

In this arena, Dr. Vasudha Murlidhar, a postdoctoral scholar in biomedical engineering at the University of California, Davis, recently developed a microfluidic device while working on her doctorate at the University of Michigan Comprehensive Cancer Center. This device can process a sample of blood and determine how many lung tumor cells are circulating in the body, thereby offering important information about the severity of the cancer and the patient's prognosis.

The device uses aspects of the cancer cell and the immune system to determine how many cancer cells are present in a blood sample. "We can engineer these devices to be able to capture cancer cells by targeting specific antigens that are present on the cells." Antigens are proteins on the surface of the cancer cells that attract cells produced by the immune system, called antibodies. "You can coat [the antigens'] corresponding antibodies on to these devices, and in a sample of blood cells that are expressing the corresponding target molecules -- in our case they would be the cancer cells -- they'll get stuck onto the device." Sort of like a magnet, the antigens and antibodies will come together, and then the pathologist can see how many cancer cells are in a known quantity of blood. This can provide important clues as to the severity of the disease and its potential for metastasizing.

[See: 7 Innovations in Cancer Therapy.]

To test the device, Murlidhar and her colleagues at the University of Michigan "spiked" cancer-free blood samples with a known quantity of cancer cells "to mimic a real sample." Once the sample "has passed through the device and the procedure for capturing the cells is done, we can go and look at them under a microscope and see how many cells were captured in the device. We know the number of cells that went in and we can count how many were captured," to determine the device's accuracy.

The device was tested on early-stage lung cancers. (Liquid biopsy technology has previously been developed in later-stage cancers, where the number of cells that are circulating is higher and thus easier to find.) Murlidhar's device attempts to provide more information and better prognostication for earlier stage patients. Although it's still in the lab and more research needs to be done, she hopes that eventually this technology will find its way into clinical settings and become part of the oncologist's arsenal of precision medical tools.

"Most of the circulating tumor cell research right now focuses more on prognostics," which can be helpful for determining the best treatment options for patients who've had a recurrence of lung cancer after having had surgery and/or chemotherapy. "I think long term, where this field would go would be diagnosis ultimately," she says.

If science can tell patients exactly how their cancer is likely to behave, that may lead to better, more precise treatments that could eliminate it all together.

Elaine K. Howley is a freelance Health reporter at U.S. News. An award-winning writer specializing in health, fitness, sports and history, her work has appeared in numerous print and online publications, including AARP.org, espnW, SWIMMER magazine and Atlas Obscura. She's also a world-record holding marathon swimmer with a passion for animals and beer. Contact her via her website: elainekhowley.com.