AI-Based Blood Test Detects Ovarian Cancer With 93% Accuracy

Ovarian cancer, often termed the silent killer, typically presents no symptoms in its initial stages, leading to late detection when treatment becomes challenging. The stark contrast in survival rates highlights the urgent need for early diagnosis: while late-stage ovarian cancer patients have a five-year survival rate of around 31% post-treatment, early detection and treatment can raise this rate to over 90%. Despite over three decades of research, developing an accurate early diagnostic test for ovarian cancer has proved challenging. This difficulty stems from the disease's molecular origins, where multiple pathways can lead to the same cancer type.

Scientists at the Georgia Tech Integrated Cancer Research Center (ICRC, Atlanta, GA, USA) have now made a breakthrough by integrating machine learning with blood metabolite information, developing a test that can detect ovarian cancer with 93% accuracy in their study group. This test outperforms existing detection methods, especially in identifying early-stage ovarian disease among women clinically considered normal. The researchers have created a novel diagnostic approach, utilizing a patient's metabolic profile to assign a more precise probability of the presence or absence of the disease.

Image: Micrograph of a mucinous ovarian tumor (Photo courtesy of National Institutes of Health)

Mass spectrometry, used to identify metabolites in blood through their mass and charge, faces a limitation: less than 7% of these metabolites in human blood have been chemically characterized. Thus, pinpointing specific molecular processes behind an individual's metabolic profile remains a challenge. Nevertheless, the team recognized the potential of using the presence of varying metabolites, as detected by mass spectrometry, to create accurate predictive models using machine learning. This approach is similar to using individual facial features for developing facial recognition algorithms.

In their innovative method, the researchers combined metabolomic profiles with machine learning classifiers, achieving 93% accuracy in a study involving 564 women from Georgia, North Carolina, Philadelphia, and Western Canada. This group included 431 active ovarian cancer patients and 133 women without the disease. Ongoing studies aim to explore the test's ability to detect very early-stage disease in symptom-free women. The vision for clinical application is a future where individuals with a metabolic profile indicating a low likelihood of cancer undergo annual monitoring, while those with scores suggesting a high probability of ovarian cancer receive more frequent monitoring or immediate referral for advanced screening.

“This personalized, probabilistic approach to cancer diagnostics is more clinically informative and accurate than traditional binary (yes/no) tests,” said John McDonald, professor emeritus in the School of Biological Sciences, founding director of the ICRC, and the study’s corresponding author. “It represents a promising new direction in the early detection of ovarian cancer, and perhaps other cancers as well.”

Related Links:
Georgia Tech