Fighting cancer better with AI

Researchers from the MEDomics consortium have got their hands on a treasure trove of bytes. One of their partners, the University of California at San Francisco (UCSF), gave them the anonymized digital medical records of 175,000 cancer patients treated between January 2010 and December 2019 at one of the health centres in UCSF. It is a huge database, which includes approximately 15 million medical notes, 200 million laboratory results, 1.7 million radiology reports, 800,000 pathology reports, 15 million drug prescriptions and three million medical reports. consultation. “This is the first time that we have had such a large volume of data, such diversity, on a wide variety of cancers and over such a long period, which has enabled us to adopt a longitudinal approach. to follow the evolution of the disease,” said the head of the MEDomics project, Oliver Morin, also head of the physics department of the department of radiation oncology of the UCSF. Once it has gone through the mill of its artificial intelligence application, the research team has thus “demonstrated the usefulness of hospital data by proving that it is possible to exploit them despite their disparity,” underlines the former engineering physics student at Laval University. Read: Montreal doc and research team receive $1 million grant for AI multiple sclerosis research The main challenge was indeed to consolidate and organize these millions of disparate pieces of information. “The data was stored in silos. Each service has its database with its own nomenclature and each health professional has its own language,” pointed out Morin. With his colleagues from McGill, Sherbrooke, Maastricht, Belgium, and Dresden, Germany universities, they spent months organizing the “data highways” that feed the algorithms of the MEDomics digital platform. Once this challenge was met, the analysis phase could begin. A predictor robot In their study published by the journal Nature Cancer on July 22, researchers show that their application, which works in machine learning, was able to update and improve a prognosis model as the disease progressed. “With this data, we were able to validate the results of published clinical studies using real data. For example, the positive impact of immunotherapy on lung cancer,” said study co-author, Dr. Catherine Park, who heads the department of radiation oncology at UCSF. “We were also able to validate several new clinical hypotheses such as the importance of cardiovascular morbidity for the results of cancer treatments or the impact of socio-demographic factors on the course of the disease and the patient’s chances of survival,” added Morin who insisted there is “enormous potential” in hospital data and, by extension, of the MEDomics application “to improve and personalize cancer treatments.” Ultimately, he imagines that it could become an easy-to-use robot, a sort of digital assistant that would help oncologists predict the course of the disease in order to adapt the therapy of their patients.

Fighting cancer better with AI

Register for free access or login.

Related Topics