AI Tool Predicts Pediatric Glioma Recurrence With Up to 89% Accuracy

An artificial intelligence (AI) tool accurately predicted the risk of relapse in patients with pediatric cancer compared with other traditional approaches, according to a recent article from The Havard Gazette.

Furthermore, the AI tool may be broadly adaptable to track and predict risk for patients with other cancers and chronic diseases undergoing surveillance imaging, according to study findings published in The New England Journal of Medicine AI.

The temporal learning model predicted recurrence of low- or high-grade glioma within one-year post-treatment with 75% to 89% accuracy, significantly outperforming single-image predictions, which have hovered at approximately 50%. Accuracy improved when more post-treatment images were used.

The AI tool is trained to analyze multiple brain scans over time, according to the news article. These results may lead to improved care for children with brain tumor gliomas, which as noted in the article, are typically treatable but vary in risk of recurrence.

“Many pediatric gliomas are curable with surgery alone, but when relapses occur, they can be devastating,” corresponding author, Dr. Benjamin Kann, said in the article. “It is very difficult to predict who may be at risk of recurrence, so patients undergo frequent follow-up with magnetic resonance 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.”

Kann is from the Artificial Intelligence in Medicine Program at Mass General Brigham and the Department of Radiation Oncology at Brigham and Women’s Hospital, located in Boston, Massachusetts. He is also an assistant professor of Radiation Oncology at Harvard Medical School.

The study included nearly 4,000 scans from 715 pediatric patients. A technique was then employed called temporal learning, which has not been used previously for medical imaging AI research, and trained the AI tool to synthesize multiple brain scans taken over the span of several months after surgery. Researchers first trained the model to sequence post-surgery MRIs chronologically, helping it learn to detect subtle changes. They then refined the model to link those changes with later cancer recurrence when appropriate.

“Many pediatric gliomas are curable with surgery alone, but when relapses occur, they can be devastating,” stated Kann in the article.

The researchers cautioned that further validation in additional settings is needed before clinical use. They want to launch clinical trials to determine whether AI-informed risk predictions can improve care by reducing imaging for low-risk patients or offering early targeted therapies to those at high risk.

“We have shown that AI is capable of effectively analyzing and making predictions from multiple images, not just single scans,” first author Divyanshu Tak of the AIM Program at Mass General Brigham and the Department of Radiation Oncology at the Brigham, said in the article. “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.”

Pediatric cancer is a term for cancers diagnosed from birth through age 14. These rare cancers differ from adult cancers in growth, spread, treatment and response. Common types include leukemia, brain and spinal cord tumors, lymphoma, neuroblastoma, Wilms tumor, retinoblastoma and cancers of the bone and soft tissue.

Most children with glioma undergo frequent brain MRIs due to unpredictable recurrence patterns, but deep learning may help improve personalized monitoring, according to the study.

Reference:

“Longitudinal Risk Prediction for Pediatric Glioma with Temporal Deep Learning” by Divyanshu Tak, et al., NEJM AI.

For more news on cancer updates, research and education, don’t forget to subscribe to CURE®’s newsletters here.