A new AI-powered blood test just proved capable of detecting 50 different types of cancers, including some that are particularly aggressive or elusive.

We know the early detection of cancer can be critical to saving lives, and scientists have long been searching for a simple and reliable blood test. While the new test is still in the early stages of development, a large trial involving 6,689 blood samples has produced some promising results.

More than 99 percent of the positive detections were accurate, meaning that there's a low chance of costly and needlessly stressful false positive diagnoses. However, there's still some risk of providing patients with false reassurance - the test detected the presence of cancer in 44 percent of cases across 50 different types of cancer.

Furthermore, detection was more sensitive the more advanced the cancer was, with 18 percent early-stage detections improving to over 90 percent if a cancer was in one of the later stages.

The test relies on a computer program trained to sift through bits of DNA that normally circulate in our blood, identifying fragments that have come from tumours. In this case the researchers focused the program on detecting how these DNA bits have been methylated - a chemical indication of whether genes are set to 'active' or 'inactive' when inside their cells.

The researchers trained the machine learning algorithm to analyse DNA methylation patterns from thousands of blood samples, before introducing it to 1,531 samples from people with cancer and 1,521 without. The program grouped these into like-patterns and the researchers then taught the AI which pattern reflected which type of cancer, before putting it to the test.

"Our previous work indicated that methylation-based tests outperform traditional DNA-sequencing approaches to detecting multiple forms of cancer in blood samples," Dana-Farber Cancer Institute oncologist Geoff Oxnard explained in a statement.

Another advantage of focusing on this epigenetic feature of methylation is that it provides clues into the types of tissues the DNA fragments come from.

"The test not only demonstrates the presence of cancer, but provides an accurate address as to the type of cancer and where the health professional should look for the malignancy," immunologist Michael Seiden told HealthDay.

The system was able to pinpoint the cancer's origins in more than 90 percent of the trial cases where cancer was detected. This information is critical for working out treatment strategies.

"Based upon this successful clinical validation in thousands of patients, the test has actually now been launched for limited use on clinical trials," Oxnard told the BBC.

However, the test still has some problems to iron out, such as its low detection rates of early-stage cancer. It also seems to have trouble identifying the origin of cancers caused by the human papillomavirus. Regardless, this study provides encouraging signs for the potential of AI-analysed blood tests.

"Before this blood test is used routinely, we will probably need to see results from clinical studies… to more fully understand the test performance," said Oxnard.

The study was published in Annals of Oncology.