Beating the pathologists: Predicting NSCLC brain metastases using AI

By analyzing lung biopsies from patients with stage I to III non-small cell lung cancer (NSCLC), artificial intelligence was able to detect individuals who will develop brain metastases and those who will not.Approximately 30% of patients diagnosed early with NSCLC will develop metastases, most commonly to the brain. That means in two thirds of cases, removal of the tumour would be sufficient to prevent recurrence and that systemic treatment wouldn't be necessary.Certain risk factors make it possible to predict with more or less accuracy the populations likely to develop metastases, but predictive biomarkers lack reliability in identifying patients at higher risk.Researchers from Washington University and the California Institute of Technology therefore joined forces and configured artificial intelligence capable of recognizing these at-risk patients. “The most difficult decision concerns stage I patients since they have the greatest chance of not developing the disease again, without resorting to systemic therapy,” note the authors of the study. Nevertheless, such treatment is still recommended in the management of these patients.Using deep learning, the researchers trained artificial intelligence on lung biopsy samples from 118 patients with stage I to III NSCLC. Over a five-year follow-up, 45 of them had developed brain metastases, and 73 were in remission.After validation of the algorithm, the artificial intelligence tool was presented with 40 new samples for which the AI didn't know the metastatic progression. In 87% of cases, the AI accurately predicted the appearance or not of brain metastases over five years. It performed especially well in recognizing cases where the cancer will not migrate out of the lung with a negative predictive value of 92.2%.These 40 cases were also referred to four independent lung pathologists who examined the tissues. They achieved an average accuracy of 57.3%, suggesting that the AI perceived details that would escape the eye of even the most experienced pathologist.What characteristics does the AI use to make its prediction?It appears that its focus is not only on the tumor cells themselves, but also on multiple features in the immediate environment of the initial tumor. Further analysis continues to identify these cellular and molecular markers detected by the algorithm, determining which are associated with possible progression of the disease.Even if these results must be validated on a larger cohort, the researchers are confident of the potential of their technology to support health professionals in their care decisions. “Identifying patients who are at risk of developing brain metastases can help us develop personalized strategies to quickly intercept cancer proliferation,” says Dr. Ramaswamy Govindan of the Washington University School of Medicine, in a press release. He added that the objective of the research was also to avoid systemic treatment, such as chemotherapy, for patients who would not need it.The researchers now want to train the algorithm to predict the progression of NSCLC to other distant organs.

Beating the pathologists: Predicting NSCLC brain metastases using AI

Register for free access or login.

Related Topics