Finding cancer in blood samples is too difficult for artificial intelligence
Blood tests for early detection of cancer have been making headlines lately. But there is still a long way to go before cancer can be detected in such a simple way.
A cancer cell divides. In doing so, it could leave traces in the blood. But because every cancer is different, we don’t know what to look for.
Symbolbild: Getty
“Blood tests will revolutionize cancer medicine!” was the headline in the German newspaper “Bild” at the beginning of June, and various Swiss media also reported this year and last year on blood tests that were touted as a breakthrough in the early detection of cancer.
In fact, numerous research groups and biotech companies around the world are working on tests that can reliably detect cancer in a blood sample. Sometimes just one type of cancer, but often several at once; one test can even detect up to fifty different types of cancer. And ideally at an early stage.
Most of these tests, better known in technical terms as liquid biopsies, look for circulating tumor DNA (ctDNA for short), proteins or other traces that enter the bloodstream when tumor cells change, grow or die. There is great hope that such tests will one day save many lives and reduce the need for expensive, stressful therapies for advanced cancer.
“We can already detect metastatic cancer in the blood quite well,” says Andreas Wicki, who, as a professor of oncology and deputy clinical director at the Clinic for Medical Oncology and Hematology at the University Hospital of Zurich, is researching the use of liquid biopsies in cancer medicine. “However, these tests are not yet sufficiently specific and sensitive to be used in screenings for the early detection of cancer.”
The question is what to look for
According to Wicki, the gap between what is possible in research and what is used in clinical practice has grown larger in recent years. “Fifteen years ago, measurement was what limited our options. Today, we can measure almost anything in a short space of time. The difficulty now is deciding what we actually want to measure – and then interpreting all the data and drawing the right conclusions from it.”
In addition, the classic algorithms used to evaluate large amounts of data are not designed for the complexity of cancer. “Machines need a lot of similar things in order to learn well.” Image recognition, for example, works very well: If you show artificial intelligence a huge database with images of melanoma and harmless moles, it can learn to recognize them.
“But if we look at the biology of cancer, it is incredibly complex. In every single patient we find countless mutations, many of which are very rare. We will never find two patients whose cancer is identical. In this variety of mutations and signals, identifying those that are the same in all patients with a certain type of cancer but do not occur in healthy people is incredibly difficult.”
A lot can be read from blood samples – but traces of cancer are too diverse.
Symbolbild: Getty
A huge hype
Viola Heinzelmann-Schwarz, professor of gynecology and chief physician of gynecological oncology at the University Hospital of Basel, sounds similar. “The hopes raised by such multi-cancer early detection tests are great; it is a huge hype,” she says. Large-scale studies are needed first to show whether, for whom and in which setting screening would be useful.
The gynecologist points out: “Even if reliable early detection in the blood is possible, it is not yet certain that screening will be of any use.” While colon or skin cancer detected early can often be cured with surgery, early detection does not improve the prognosis for all types of cancer. A few years ago, for example, a British study with more than 200,000 women showedthat screening with transvaginal ultrasound and blood tumor markers made it possible to detect ovarian and fallopian tube cancer in stages one and two.
Of the approximately 100,000 women who underwent cancer screening every year for an average of 16 years, however, not significantly fewer died in the end than in the control group of 100,000 women who did not take part in the screening. Therefore, the study authors did not recommend screening the general population for ovarian and fallopian tube cancer.
Heinzelmann-Schwarz sees great potential for liquid biopsies – but not for the time being in the early detection of cancer, but rather to better treat patients in whom cancer has already been diagnosed. In these cases, they can be used as one of many methods to create an in-depth biological profile of the tumor, which helps to select the best therapy in each case. Anreas Wicki shares this view: “At the moment, liquid biopsies are primarily a useful tool in medical practice to find out whether we are successfully treating a tumor.”
Thibaud Kössler, oncologist at the University Hospital of Geneva and senior researcher at the Gastrointestinal Cancer Research Lab, is already using liquid biopsies to better treat patients with metastatic cancer. “Liquid biopsies do not yet provide us with information as precise as a biopsy of cancerous tissue. However, they are much less invasive and can also be used when the tumor or metastases are located in such a way that it is difficult or even dangerous to take tissue samples. In the future, we will use them regularly to monitor how well the therapy is responding.”
Although he also believes that a lot of research is still needed before they can be used as a screening tool, Kössler assumes that the US Food and Drug Administration (FDA) will approve the first blood-based tests for early cancer detection in the near future. “Research is almost there. The medical associations must keep an eye on the topic and develop appropriate guidelines. Because it is only a matter of time before the first patients in the USA order such tests – and then visit their GP with a test result that is difficult to classify.”
