"Many pediatric gliomas are curable with surgery alone, but when relapses occur, they can be devastating," corresponding author, Benjamin Kann, MD, of the Artificial Intelligence in Medicine (AIM) Program at Mass General Brigham and the Department of Radiation Oncology at Brigham and Women's Hospital, said in a press release.
"It is very difficult to predict who may be at risk of recurrence, so patients undergo frequent follow-up with magnetic resonance (MR) imaging for many years, a process that can be stressful and burdensome for children and families. We need better tools to identify early which patients are at the highest risk of recurrence," he added.
For the study, researchers developed a self-supervised temporal deep learning approach in which a multistep model encodes patients' serial MRI scans and is trained to classify the correct chronological order. They fine-tuned the model to predict 1-year recurrence of pediatric gliomas using patients' historical postoperative surveillance scans. The model was applied to 3,994 scans from 715 patients for low and high-grade gliomas.
Ultimately, the model predicted recurrence of low and high-grade glioma at one-year post-treatment with 75- 89% accuracy, a significant improvement on the 50% accuracy from single images alone. Accuracy plateaued at four to six images.
More validation across additional settings is required for the technology to be used in clinical contexts, cautioned the researchers. They hope to launch clinical trials to find out whether AI-informed risk prediction can improve care.
"This technique may be applied in many settings where patients get serial, longitudinal imaging, and we're excited to see what this project will inspire," first author Divyanshu Tak, MS, of the AIM Program at Mass General Brigham and the Department of Radiation Oncology at the Brigham, said in a press release.
Sources: Science Daily, The New England Journal of Medicine AI
