Photo credit: www.sciencedaily.com
The potential of artificial intelligence (AI) in the realm of medical imaging is becoming increasingly apparent, particularly in the analysis of extensive datasets that may reveal significant insights often overlooked by human evaluators. A recent study indicates that AI-assisted evaluation of brain scans can enhance the management of pediatric patients suffering from gliomas, a type of brain tumor that is generally treatable but carries varying risks of recurrence. Researchers from Mass General Brigham, alongside teams from Boston Children’s Hospital and Dana-Farber/Boston Children’s Cancer and Blood Disorders Center, have focused on developing deep learning algorithms designed to analyze follow-up brain scans after treatment, aiming to identify patients who are at heightened risk of tumor recurrence. Their findings have been detailed in The New England Journal of Medicine AI.
“While many pediatric gliomas can be effectively treated through surgical intervention alone, the consequences of recurrence can be severe,” remarked Benjamin Kann, MD, a leading author involved in the study from the Artificial Intelligence in Medicine (AIM) Program at Mass General Brigham and the Department of Radiation Oncology at Brigham and Women’s Hospital. “Predicting which patients are vulnerable to relapse is extremely challenging, leading to frequent, anxiety-inducing follow-ups with magnetic resonance imaging (MRI) over many years. We urgently require more effective tools to promptly identify those at the highest risk of recurrence.”
Research into infrequent conditions, such as childhood cancers, often encounters difficulties due to limited data availability. In this study, partially funded by the National Institutes of Health (NIH), investigators utilized collaborative relationships across various institutions to compile close to 4,000 MRI scans from a cohort of 715 pediatric patients. To enhance the AI’s learning capabilities from patients’ brain scans and improve the accuracy of recurrence predictions, the researchers adopted a method known as temporal learning, which trains algorithms to analyze serial images acquired over time following surgery.
Conventional AI models for medical imaging typically assess individual scans. However, this study introduced temporal learning to facilitate predictions based on the evaluation of images collected at different time intervals. Initially, the researchers guided the model to organize a patient’s MRI scans in chronological order, allowing it to discern subtle variations over time. Subsequently, they refined the model to correctly correlate these changes with incidents of cancer recurrence when applicable.
The findings revealed that the temporal learning model could predict the recurrence of gliomas—whether of low or high grade—within a year post-treatment with an impressive accuracy range of 75-89 percent. This marked a significant improvement compared to predictions derived from single images, which only achieved about 50 percent accuracy, akin to random guessing. It was noted that as the model received images from additional time points post-treatment, its predictive accuracy improved, although enhancements plateaued after four to six images.
The research team emphasized the necessity for further validation in different clinical settings before implementing these advancements in practice. Ideally, they aspire to initiate clinical trials to examine whether AI-derived risk assessments can enhance patient care outcomes—potentially by reducing imaging frequency for those deemed at low risk or by proactively administering targeted therapies to patients identified as high risk.
“Our study demonstrates that AI possesses the capability to analyze and generate predictions from multiple scans over time rather than relying solely on individual images,” stated Divyanshu Tak, MS, a co-author from the AIM Program at Mass General Brigham and the Department of Radiation Oncology at Brigham and Women’s Hospital. “This innovative technique holds promise for various situations involving patients who undergo continuous, longitudinal imaging, and we look forward to the future applications it could inspire.”
Authorship: Along with Kann and Tak, the research team from Mass General Brigham includes Biniam A. Garomsa, Anna Zapaishchykova, Zezhong Ye, Maryam Mahootiha, Tafadzwa Chaunzwa, Hugo JWL Aerts, and Daphne Haas-Kogan. Other contributors comprise Sridhar Vajapeyam, Juan Carlos Climent Pardo, Ceilidh Smith, Ariana M. Familiar, Kevin X. Liu, Sanjay Prabhu, Pratiti Bandopadhayay, Ali Nabavizadeh, Sabine Mueller, and Tina Y. Poussaint.
Funding: This research received partial support from the National Institute of Health/National Cancer Institute (NIH/NCI) under grant numbers U54 CA274516 and P50 CA165962, alongside contributions from the Botha-Chan Low Grade Glioma Consortium. Acknowledgment is also extended to the Children’s Brain Tumor Network (CBTN) for providing access to imaging and clinical data.
Source
www.sciencedaily.com