THANKS. Listen to this article using the player above. ✖
Do you want to listen to this article for free?
Fill out the form below to unlock access to all audio items.
Artificial intelligence (AI) shows a huge promise to analyze large sets of medical imaging data and identify the models that can be missed by human observers. Assisted interpretation of the AI of brain scintigraphs can help improve care for children with brain tumors called gliomas, which are generally treatable but vary at risk of recurrence. Investigators of Mass general Brigham And employees of the Boston Children’s Hospital and Dana-Farber / Boston Children’s Cancer and Blood Disorders Center have formed in depth learning algorithms to analyze sequential and post-processing brain analyzes and the reporting of patients at risk of recurrence of cancer. Their results are published in The New England Journal of Medicine Ai.
“Many pediatric gliomas are curable with surgery alone, but when relapses occur, they can be devastating,” said the corresponding author Benjamin Kann, MDof the Artificial Intelligence Program in Medicine (AIM) to general of mass Brigham and the ONCOLOGY RADIATION DEPARTMENT has Brigham and Women’s Hospital. “It is very difficult to predict that can be at risk of recurrence, so patients are in common with magnetic resonance imaging (MR) for many years, a process that can be stressful and restrictive for children and families. We need better tools to identify early which patients present the highest risk of recurrence. ”
Relatively rare disease studies, such as pediatric cancers, can be disputed by limited data. This study, which was partly funded by the National Institutes of Health, operated institutional partnerships across the country to collect nearly 4,000 MR analyzes with 715 pediatric patients. To maximize what AI could “learn” from a patient’s brain scanners – and predict recurrence more precisely – researchers used a technique called time learning, which forms the model to synthesize the results of several brain scans supported by several months after surgery.
As a rule, AI models for medical imaging are formed to draw conclusions from unique scans; With temporal learning, which has not been used before for medical imaging, the images acquired over time shed light on the prediction by the algorithm of the recurrence of cancer. To develop the temporal learning model, the researchers first formed the model to sequence the analyzes of a post-operative MR of a patient in chronological order so that the model can learn to recognize subtle changes. From there, the researchers refined the model to properly associate the changes with the recurrence of subsequent cancer, if necessary.
In the end, the researchers found that the temporal learning model predicted the recurrence of a low or high grade glioma by one year after treatment, with a precision of 75 to 89% – significantly better than the precision associated with predictions based on unique images, which they considered about 50% (no better than chance). Providing AI with images of more time after treatment increased the precision of the prediction of the model, but only four to six images were necessary before this improvement is composed.
Researchers warn that additional validation in additional parameters is necessary before clinical application. In the end, they hope to launch clinical trials to see if AI -focused risk forecasts can lead to improvements in care – whether by reducing imagery frequency for low -risk patients or preventing high -risk patients with high -risk patients with targeted adjuvant therapies.
“We have shown that AI is capable of analyzing effectively and making predictions from multiple images, not just unique analyzes,” said the first author Divyanshu tak, msof the AIM program of the mass general Brigham and the Ministry of Retlaim ONCOLOGY in Brigham. “This technique can be applied in many contexts where patients obtain longitudinal imaging in series, and we are delighted to see what this project will inspire.”
Reference: Tak D, Garomsa Ba, Zapaishchykova A, et al. Prediction of the longitudinal risk for pediatric glioma with temporal depth learning. Nejm ai. 2025; 2 (5): AIOA2400703. DOI: 10.1056 / AIOA2400703
This article was republished from the following materials. Note: The material may have been modified for duration and content. For more information, please contact the cited source. Our press release policy is accessible here.
