AI Blood Test Shows Promise for Early Ovarian Cancer Detection

Ovarian cancer kills quietly. Women often feel nothing until the disease has already spread, which is why it remains the fifth leading cause of cancer death in the United States. Researchers at Johns Hopkins have been chasing a solution to this silence for years—a blood test sensitive enough to catch the cancer early, before symptoms arrive. This week, they presented evidence that they may have found it.

The test works by hunting for fragments of tumor DNA circulating in the bloodstream. Cancer cells grow and die rapidly, leaving behind genetic debris with a distinctive pattern—chaotic and different from the DNA fragments shed by healthy cells. A team led by Dr. Victor Velculescu, co-director of the Cancer Genetics and Epigenetics Program at Johns Hopkins Kimmel Cancer Center, trained an artificial intelligence system to recognize these patterns. The AI didn't work alone. It also factored in measurements of two known ovarian cancer markers, proteins called CA125 and HE4, combining all three signals into a single assessment.

The results, presented Tuesday at the American Association for Cancer Research annual meeting in San Diego, were striking. The test showed almost no false positives—a specificity of 99 percent, meaning women without cancer were almost never told they had it. But the real power lay in how it performed across different stages of disease. For stage 1 cancers, the earliest and most treatable form, the test caught the disease 69 percent of the time. That rate climbed to 76 percent for stage 2, 85 percent for stage 3, and reached 100 percent for stage 4 cancers. These numbers substantially outpaced what researchers could achieve using CA125 measurements alone, the current standard approach.

The study itself was modest in scale—134 women with ovarian cancer, 204 without cancer, and 203 with benign ovarian masses. Jamie Medina, a postdoctoral fellow and co-first author, explained the underlying logic in straightforward terms: by analyzing DNA fragments across the entire human genome, researchers could detect the subtle fingerprints of malignancy that a single protein marker might miss. The AI learned to see patterns that human analysis alone could not reliably identify.

Velculescu emphasized that these findings remain preliminary. They were presented at a medical conference, not yet published in a peer-reviewed journal, which means the scientific community has not yet subjected them to the full scrutiny that publication demands. Larger studies will be needed to confirm whether these detection rates hold up in different populations and settings. But his tone suggested confidence in the direction. He framed this work as part of a larger body of research from his group demonstrating that genome-wide analysis of cell-free DNA fragments, combined with machine learning, can detect cancers with high accuracy.

The stakes of this research are substantial. Ovarian cancer typically produces no symptoms in its early stages, when treatment is most likely to succeed. By the time women notice something wrong—abdominal bloating, pain, changes in appetite—the cancer has often advanced. A blood test that could catch the disease before symptoms appear would represent a fundamental shift in how the disease is managed. It would not be a cure, but it could be the difference between catching cancer when it is still confined to the ovary and catching it after it has spread throughout the abdomen. For now, the test remains a research tool. The next step is the harder one: proving that it works reliably in the real world, in larger numbers of women, before it can move from the laboratory into clinical practice.