AI test could be first screening tool for ovarian cancer

Machine learning has been used to develop a promising screening test for ovarian cancer, a team of US-based researchers has said.

The researchers, based at the Dana-Farber Cancer Institute and the Brigham and Women's Hospital in Boston, used tiny RNA fragments called microRNAs as the basis for the artificial intelligence test, publishing their results in eLife. MicroRNAs are involved in gene activation and circulate in the blood. They act as the copyeditors of the genome.

'Before a gene gets transcribed into a protein, [microRNAs] modify the message, adding proofreading notes to the genome,' said Dr Kevin Elias, from Brigham and Women's Hospital's department of obstetrics and gynecology, lead author of the study.

The team gathered information on microRNAs in blood samples from 135 women before they had surgery or chemotherapy. They used this to train a computer program to differentiate cases of ovarian cancer from benign tumours and healthy tissue.

This program was first tested on data from 859 women with and without ovarian cancer. The sensitivity (not missing any cancers) and specificity (not flagging up healthy people) were better than the current tests for ovarian cancer.

The test was then used to predict the diagnoses of 51 patients in Poland. The test gave results with 91.3 percent sensitivity and 78.6 percent specificity. In other words, if the test says you have cancer, there is around a 1 in 10 chance that it's wrong; if the test says you are healthy, there is a 1 in 5 chance that it's wrong. This level of accuracy is similar to that of a Pap smear test for cervical cancer.

At the moment, ovarian cancer is typically diagnosed late, which is one of the main reasons for patients' poor survival rate. No FDA-approved screening techniques exist for ovarian cancer. Current early-detection tests have a high false positive rate, resulting in many women being mistakenly told they have cancer when they don't. These tests, such as ultrasound or detection of the protein CA125, are so ineffective that clinical trials have indicated that they have no effect on survival rates.

'The key is that this test is very unlikely to misdiagnose ovarian cancer and give a positive signal when there is no malignant tumour. This is the hallmark of an effective diagnostic test,' said senior author Dr Dipanjan Chowdhury, chief of the division of radiation and genomic stability in the department of radiation oncology at Dana-Farber.

The team will now plan a long-term study following a group of healthy women, to see how their microRNAs change as some develop ovarian cancer. The scientists hope that the test could be used for general population screening, as well as for women at high risk of ovarian cancer.