In Paris, the Curie Institute has led the creation of artificial intelligence capable of tracing the origins of particularly devious cancers, with a very high success rate. An instrument that could make it possible to better tailor treatments, thus increasing the chance of survival of thousands of French people every year.
The place doesn’t look all that special, halfway between a very classic office and a laboratory – labeled vials covering the work surfaces, a few analysis machines standing quietly on the throne. However, it is here, at the Institut Curie (Paris), that a potentially revolutionary discovery was born for thousands of patients: artificial intelligence capable of detecting the origins of certain cancers with much greater precision and speed than human specialists.
This promising AI is largely Doctor Sarah Watson’s baby. This double-hatted specialist is both a doctor (medical oncologist) and a researcher at the Curie Institute and mainly works on rare tumors. And among the research topics there is one that still poses major problems for specialists today: “cancers of unknown primary” (CPI).
Common protocols that “work very poorly”
“CPIs are cancers that manifest at a metastatic stage, that is, there are lesions in different places in the body, but we do not know where these metastases come from,” explains Tech&Co Sarah Watson. “The problem is that today in oncology we don’t initially treat cancer in the same way in the breast, colon, kidneys, etc..”
So without knowing the origin of the cancer, it is impossible to treat it accurately – and effectiveness suffers. “We treat these patients with chemotherapy protocols that are not specific at all, and that generally work very poorly,” complains Sarah Watson, and “we spend a lot of time identifying the primary tumor” (time that could be used to cured). ).
“Patients (affected by this type of cancer) have a life expectancy of less than a year from diagnosis,” Sarah Watson points out.
To improve care for these ICCs (which affect 6,000 to 7,000 patients per year in France), Sarah Watson decided to use an increasingly popular method: artificial intelligence.
“Tripled survival time”
Like most technological developments, it starts from a hypothesis. “We told ourselves that perhaps the molecular signatures of these metastases retained some features of their tissues of origin,” says Sarah Watson. In summary, do the metastases spread throughout the body leave a trace on their genetic ‘identity map’ that is specific to their area of origin – which could make it possible to localize it and thus treat the cancer at its source?
To find out, the Institut Curie, in collaboration with specialists from several other centers in France, has trained software that focuses on a single task: recognizing the origin of cancers from the RNA sequences of tumors identified by biopsy found. Firstly, by allowing it to register more than 20,000 different samples of known origin, so that it can obtain references for analysis.
The professionals then submitted tumor samples to the AI, without telling them what their origin was, to see if it was able to give the correct answer. Then came the fateful step: “Now that she knows how to recognize the different types of cancer, if we present her with a CPI tumor, will she be able to compare it to well-identified cancers?”
The AI was tested using standard tools on about fifty tumors of unknown origin. “In more than 80% of cases, the AI was able to tell us the tissue of origin with extremely high prediction scores,” explains Sarah Watson from Tech&Co.
“Patients for whom we can identify the tissue of origin have a threefold (duration of) survival compared to patients in the opposite case,” rejoices Sarah Watson of Tech&Co.
“About ten patients” suffering from CPI could even benefit from a treatment adapted to the origin proposed by AI. “We have had very promising signals,” says Sarah Watson, with even “one patient achieving a complete response thanks to a very specific treatment.”
“A very small series of patients,” Sarah Watson emphasizes, but that gives hope for big things. Enough to integrate AI into routine clinical care at the Institut Curie in the case of CPI, and expand to the whole of France as part of a sequencing plan. “Several dozen patients have already had their profiles analyzed by this tool.”
“AI is still a tool”
Sarah Watson wants to continue improving her tool. “In 20% of cases, the AI was unable to make a reliable diagnosis with sufficient confidence scores,” says Sarah Watson. “We have two possible explanations: an AI tool can only recognize what it has learned to recognize (so more data needs to be added), and we are working on very small samples that could have been damaged or poorly preserved,” which their analysis makes it more difficult.
The specialist also wants to spread it even more widely and perfect it by teaching her to analyze other types of data: DNA profile, digitized anatomopathological slides, etc. All with quality data, representative of the diversity of patients and respecting their consent. insists.
“It is more complex, but it has already been done for other types of pathologies,” Sarah Watson assures Tech&Co.
Until a tool is created that can replace specialists? For Sarah Watson, the question does not arise: “The tool not only works, but can be integrated with a series of data analyzed during a meeting of experts to make the diagnosis as accurate as possible.
“AI should not be scary, but should be seen as an additional tool: we had the stethoscope, the microscope, the MRI…,” sums up Sarah Watson. “AI is another tool for analyzing the data we are confronted with that is not necessarily useful to the human brain.”
“Analyze the results that AI produces, know how to explain them to patients and make a therapeutic decision…” Sarah Watson assures: “The computer will never work, it will remain in the hands of the doctors.”