Artificial intelligence is destined to revolutionize many fields of science, starting with medicine. AI can in fact help doctors to personalize therapies, provide more precise and timely diagnoses, predict which patients will respond best to drugs, and thus choose the therapeutic strategy that has the highest chance of success. An excellent example of what awaits us in the coming years comes from a study carried out by researchers from the University of Cambridge and the Catholic University of the Sacred Heart, published in recent days on Nature Communicationswhich tested the ability of AI to predict how ovarian tumors will respond to therapies, obtaining accurate predictions in 80% of patients.
A difficult tumor
Ovarian cancer is a complex tumor. In the initial stages it almost always presents non-specific symptoms that complicate diagnoses, and for this reason patients almost always discover the disease when it is in advanced stages, and is therefore more difficult to treat. The most common form of this neoplasm, high-grade ovarian serous carcinoma, is also an aggressive disease that is frequently resistant to chemotherapy. The lack of markers or predictive tools means that only in about 50% of cases doctors are able to predict the evolution of the disease and the tumor’s response to pharmacological therapies. And this means that oncologists have difficulty deciding which patients to undergo neoadjuvant therapies, which can reduce the tumor mass before the operation to simplify and make the operation less invasive, and which are more likely to benefit from an immediate appeal to surgery.
The AI will take care of it
To look for a strategy with which to improve the ability to predict the response to chemotherapy, the team of researchers led by Evis Sala, professor of radiotherapy at the Catholic University of the Sacred Heart, decided to rely on artificial intelligence. The program was trained using two datasets relating to 134 patients, which included clinical characteristics, information on the therapies performed, the presence of biomarkers and circulating tumor DNA (ctDNA), and quantitative characteristics of the tumors obtained with computed tomography.
The model, which the researchers called IRON (Integrated Radiogenomics for Ovarian Neoadjuvant therapy), made it possible to stratify the probability of success of noadjuvant chemotherapy based on different characteristics of the patients and their tumors. Tested on a new dataset, it demonstrated that it can accurately predict the probability of success of therapies in approximately 80% of patients. Obviously, the results are still preliminary for now, and will require a clinical trial before Iron can actually be used to help patients with ovarian cancer. Fortunately, it shouldn’t take too long.
“From a clinical point of view, the framework we have proposed responds to the need to early identify patients who are unlikely to respond to neoadjuvant therapy, and for whom it is better to proceed immediately with surgery,” explains Sala. “Our tool can be used to stratify the risk of each patient, and this is what we will do in a research that we are carrying out at the Gemelli Polyclinic in collaboration with the team of Giovanni Scambia, Professor of gynecology and obstetrics at the Università Cattolica del Sacro Heart and Scientific Director of the Agostino Gemelli University Polyclinic Foundation”.