Blood test using AI shows promise for early ovarian cancer detection

More than 300,000 women a year receive an ovarian cancer diagnosis, most of them over fifty. By the time they learn they have the disease, it has often already advanced—a cruel arithmetic that makes treatment harder and outcomes worse. The symptoms that might have signaled something was wrong—bloating, pelvic pain, feeling full quickly after eating—are easy to miss or dismiss as something ordinary. A woman might wait months before a doctor orders the scans and biopsies that finally confirm what's happening.

Now researchers from universities in Manchester and Colorado have developed a blood test that could change that timeline. The test works by hunting for what ovarian cancer leaves behind in the bloodstream: fragments released by cancer cells that carry lipids—tiny, fat-like molecules—alongside certain proteins. These two things together form what amounts to a biological signature unique to ovarian cancer. A machine learning algorithm, trained on thousands of patient samples, has learned to spot the patterns in this signature that a human eye would miss.

The researchers tested the platform on 832 blood samples. At the University of Colorado site, the test identified ovarian cancer with 93% accuracy across all stages of the disease and 91% accuracy when the cancer was caught early. At Manchester, the numbers were slightly lower but still substantial: 92% accuracy overall and 88% in early stages. These results were published in Cancer Research Communications, a journal of the American Association of Cancer Research.

The current diagnostic pathway for ovarian cancer is a patchwork—ultrasound scans, CT scans, needle biopsies, laparoscopy, sometimes surgery itself. Each step takes time. Each step carries its own burden. A simple blood test that could narrow the field, that could tell a doctor "look harder here" before symptoms have metastasized into late-stage disease, would reshape how the disease is caught.

Alex Fisher, the chief operating officer and co-founder of AOA Dx, the company that developed the test, said it can detect ovarian cancer "at early stages and with greater accuracy than current tools." Abigail McElhinny, the chief science officer at AOA Dx, framed it differently: by combining multiple types of biomarkers through machine learning, they had built a tool that could navigate the molecular complexity of ovarian cancer across its different subtypes and stages. She suggested the platform could improve early diagnosis and, as a consequence, lower costs to healthcare systems while improving outcomes for patients.

Emma Crosbie, a professor at the University of Manchester and an honorary consultant in gynaecological oncology at Manchester University NHS Foundation Trust, said the platform "has the potential to significantly improve patient care and outcomes." She and her colleagues plan to move forward with prospective trials—studies that follow patients forward in time rather than backward through existing data—to validate the test further and understand how it might fit into the systems already in place in hospitals and clinics.

What remains to be seen is whether a test that works well in a research setting will work as well when deployed in the real world, where blood samples come from different labs, where patient populations vary, where the pressures of routine clinical practice are different from the controlled conditions of a study. The researchers have named this as their next task. The promise is real. The work is far from finished.