When artificial intelligence identifies the unknown origin of metastasized cancer

by time news

January 2020. The case of a 30 year old young man presenting “metastases all over the place”, remembers Sarah Watson, is entrusted to the laboratory of this biologist and oncologist from the Institut Curie. “We expected a diagnosis of sarcoma [cancer rare des tissus mous ou de l’os], which is my specialty. But the biopsy shows us that it’s a carcinoma [cancer d’un tissu épithélial]she explains. We didn’t know what the primitive was [premier organe touché]. »

The prognosis is not good. Most cancers are now treated in a targeted manner depending on the organ of origin. Those, like this one, called “unknown primitive”, “must be treated according to international recommendations, with the widest possible chemotherapy that can hit a little on all possible origins, explains the oncologist. But this treatment works very poorly overall, and these patients have a life expectancy of between six and ten months..

Read also: Article reserved for our subscribers How artificial intelligence is shaking up medical practices

Fortunately, the scientist and her team have just developed an artificial intelligence (AI) tool for these cancers. Between May and December 2019, a learning algorithm, designed by a laboratory researcher, was trained on a database made up of more than 20,000 RNA (ribonucleic acid) profiles of colon cancer tumor cells, breast, lung… So many identity cards for a given tumour, RNA being the expression signature of the genes associated with it. “The principle of an AI tool is to be able to recognize something that it has already encountered, recalls the researcher. We said to ourselves: if we manage to teach a computer to distinguish, on the level of RNA, a tumor of the kidney, of the colon or of the breast, perhaps this tool will be capable, if we submit to it the RNA from a cancer of unknown origin, to find where it came from. » The result was not guaranteed: “Cancers of unknown origin perhaps had their own RNA profile? »she adds.

The “first patient”

A l’Institut Curie, “the young man was not well at all, she remembers. Without really knowing if it would work, we decided to test our tool. » Getting the tumor RNA profile takes a week. Computer processing takes a few minutes. As a result, the young man has kidney cancer “with 95% certainty according to the software”, she specifies. This patient will therefore be treated specifically for the kidney, a protocol without chemotherapy, because this organ is not very receptive to it.

You have 41.21% of this article left to read. The following is for subscribers only.

You may also like

Leave a Comment