Researchers from the NYU Grossman School of Medicine and Perlmutter Cancer Center have successfully trained an artificial intelligence (A.I) tool to predict the response to immunotherapy in patients with metastatic melanoma. With an accuracy rate of 80%, it is hoped that the tool can be implemented into clinical practice once an accuracy of 90% is achieved.
“Our findings reveal that artificial intelligence is a quick and easy method of predict how well a melanoma patient will respond to immunotherapy,” says Paul Johannet, MD, a postdoctoral fellow at NYU Langone Health and the Perlmutter Cancer Center, as well as the study’s first author.
Recently published in Clinical Cancer Research, the study collected 302 images of tumour tissue samples from 121 patients who had been treated with immune checkpoint inhibitors at the NYU Langone hospital. These images were divided in 1.2 million pixels, which were then given to the A.I program, along with other factors, such as the severity of disease, which kind of immunotherapeutic drug they had been given and how the patient responded to that particular treatment. Through an in silico ‘trial-and-error’ process known as deep learning, the program was eventually able to predict patient outcomes with a clinically significant degree of accuracy. These results were further validated with an additional 40 images from 30 patients at Vanderbilt University in Nashville, which used different imaging equipment and sampling techniques.
The utilisation of neural networks and deep learning as a diagnostic or prognostic tool is an emerging area of research. In the context of metastatic melanoma, a cancer with particularly poor prognostic outcomes, this tool will allow physicians to select the most appropriate treatment for each patient. Not only will this tool give the best chance of survival to each individual patient, it will save patients from the potential side effects of inappropriate treatment. Perhaps as significant as the development itself, is the fact that this system requires no special or additional hardware- the program can be run on a standard computer and utilises histological slide scanners that are commonplace in a clinical setting. Ultimately, this means that this technology could be seamlessly integrated into clinical practice without addition cost.
In 2018 alone, 3489 cases of metastatic melanoma were diagnosed in Belgium, with this number having increased progressively since 2009, in which 1913 cases were reported. With a clear unmet clinical need, this trend could potentially be mitigated, and even reversed, with the implementation of this prognostic tool into clinical practice.
Johannet. P et al., Using Machine Learning Algorithms to Predict Immunotherapy Response in Patients with Advanced Melanoma. Clinical Cancer Research 2020.