When a patient is diagnosed with cancer, a period of fear and confusion inevitably follows. The patient and their doctor may share similar questions: What are the next steps? Which treatment should be pursued? What does the future hold? 

With the use of deep learning, artificial intelligence can help address these difficult questions for doctors and patients, and may ultimately help determine a course of action and treatment options, according to the recent study, “Survival Analysis of Localized Prostate Cancer with Deep Learning.” 

In collaboration with the Department of Veterans Affairs and the Department of Energy, a group of researchers at Brookhaven National Laboratory used a variety of patient health histories to develop deep learning models to predict localized prostate cancer prognoses at the 2-year, 5-year, and 10-year marks. Subgroup analyses were also performed based on race and age. 

Dr. Xin Dai, Research Associate at Brookhaven National Laboratory, provides an overview of the process. 

“Given the patient’s history up to ten years before the diagnosis, we utilize different variables like their ages, previous lab tests, races –all kinds of information–and put it into a deep learning model,” says Dai. “Then the model will give out the predictions of their survival rates.” 

“Based on a discussion I had with a medical doctor from the Department of Veterans Affairs, what doctors want is not just a prediction, but more so a survival analysis,” says Dr. Shinjae Yoo, Computational Scientist at Brookhaven National Laboratory, and Adjunct Assistant Professor at Stony Brook University. “That’s why we focus on this.” 

Additionally, these models have a unique and useful asset: uncertainty quantification. This feature reveals how confident the models are in their predictions. 

This study reveals AI’s great potential in the medical field, specifically in clinical decision making. With AI, doctors can be guided towards what’s right for the patient based on their survival analysis. 

Any potential collaborators interested in the code or hearing more about the study may contact Dr. Yoo [sjyoo@bnl.gov] or Dr. Dai [xdai@bnl.gov]. 

-Sara Giarnieri, Communications Assistant