Improving ovarian cancer detection with AI
About 75% of ovarian cancer patients are not diagnosed until the disease has already spread to other parts of the body, leading to a poor treatment outlook. There is no reliable way to screen for ovarian cancer, and its symptoms are vague and easy to mistake for less serious problems.
UW Carbone Cancer Center researcher Irene Ong, PhD, is using artificial intelligence to search patients’ electronic health records for early warning signs of ovarian cancer. Her team compares health records from patients before they were diagnosed with records from other patients, looking for shared biological patterns that could serve as an early warning.
“So often ovarian cancer is not caught until it’s stage III or IV, when the cancer has metastasized, and their five-year survival is 40% or less,” Ong explained. “But if you catch it early, before it spreads, then there's 90% or greater five-year survival rate.”
People are also reading…
Ong will be one of three UW Carbone researchers discussing AI’s potential to transform cancer care during the Cap Times Idea Fest on Friday, Sept. 25, at Memorial Union. Reserve your tickets for this free event at captimesideafest.ticketsauce.com.
Ong is an assistant professor of obstetrics and gynecology and biostatistics and medical informatics at the UW School of Medicine and Public Health. Her lab builds AI models that use biological and medical data on a range of cancer types. Her projects include improving cancer screening and early detection, as well as developing methods to study cancer cells at the molecular level to guide targeted treatment research.
Why ovarian cancer is so difficult to detect
A woman’s lifetime risk of developing ovarian cancer is about 1 in 91, and the median age of diagnosis is 63. Symptoms include abdominal bloating, pain in the belly and pelvis, and digestive trouble, and they are often mistaken for other, more common health problems.
When a doctor suspects a woman has ovarian cancer, they perform a pelvic exam, order imaging of the abdomen and pelvis, and run a blood test that checks for high levels of a protein called Cancer Antigen 125.
“While the CA125 test has high sensitivity and can detect ovarian cancer, it has low specificity,” Ong said of the blood test. “This means it often flags the presence of abnormal proteins but struggles to differentiate between benign inflammation and malignant cancer in a healthy person.”
Training AI to spot early warning signs
For the past four years, Ong’s lab has been training an AI model with deidentified patient data from UW Health and the National Institutes of Health’s All of Us Research Program. The team compares health records from ovarian cancer patients that was gathered before their diagnosis, with records from patients who had similar symptoms but no ovarian cancer diagnosis, patients diagnosed with a different type of cancer, and other patient groups. Up to 20% of ovarian cancers are tied to an inherited genetic risk factor, and the team is also studying that link using the All of Us database.
Building better screening tools
With these large pools of data, the team is trying to identify reliable biological signs of early ovarian cancer development. Ong envisions a clinical tool that could help doctors and patients monitor risk and catch the disease early, when it is most treatable.
“AI tools can be useful for providing information to the clinicians and having them include that data in their thinking and evaluation,” she said. “It can assist the clinicians, rather than having the AI just make the decision.”
Beyond ovarian cancer
Ong said this data-driven AI approach could also help identify early warning signs of other cancers. One example is pancreatic cancer, which can involve digestive system symptoms similar to ovarian cancer and is also often caught at an advanced stage.
The future of cancer biomarkers
Ong also hopes to expand this research to include blood and tissue samples to find biological markers. The goal is to develop a blood test for ovarian cancer warning signs, or biomarkers that might better inform clinical decisions. That could give doctors another tool for monitoring at-risk patients, and a lower-risk method to screen suspected ovarian cancer patients before diagnostic surgery.
