In a recent interview, Benjamin Kann, M.D., discussed the use of an emerging AI model that can assess longitudinal MR imaging to help predict postoperative recurrence risk for gliomas in pediatric patients.
There is significant variability with respect to predicting postoperative recurrence risk in pediatric patients who undergo surgery for gliomas, which contributes to considerable burden and stress for patients and their families. With this in mind, Benjamin Kann, M.D., and his colleagues developed a temporal deep learning model that evaluates longitudinal magnetic resonance imaging (MRI) to ascertain the risk of pediatric glioma recurrence.
In a new study, published earlier today in New England Journal of Medicine AI, Dr. Kann and his colleagues found that the model’s average area under the receiver operating characteristic curve (AUROC) for predicting one-year event-free survival (EFS) with longitudinal MRI assessment was 80.6 percent in contrast to 50.3 percent for single MRI scan assessment.
In a recent interview, Dr. Kann said the temporal deep learning model may provide a viable option for synthesizing current and previous imaging at the point of care.
“The point is to really take that historical imaging and analyze it all together to provide the most accurate prediction of… what will happen within the next year for this patient,” added Dr. Kann, an assistant professor of radiation oncology at Harvard Medical School.
While emphasizing that subsequent prospective research will be necessary to validate the study findings, Dr. Kann noted the potential of the temporal deep learning model in facilitating earlier treatment interventions for patients deemed to have high recurrence risk and perhaps longer intervals between follow-up MRI scans for patients identified as having a lower risk of glioma recurrence.
Dr. Kann also suggested broader implications with the study perhaps serving as a springboard to more investigations with AI assessment of longitudinal imaging versus single scans.
“I think there's been a lot done at single time point analyzes, and for a long time, it's been seen as kind of unreachable or infeasible to do these types of longitudinal medical imaging analyzes. … We hope that we've shown that there is a path forward there, and (we’re) just excited to see other groups do even more interesting and innovative things inspired by our work,” noted Dr. Kann, a radiation oncologist affiliated with the Brigham and Women’s Hospital and the Dana-Farber Cancer Institute in Boston.
(Editor’s note: For related content, see “Emerging PET Agent Gets FDA Fast Track Designation for Glioma Imaging,” “A Closer Look at MRI-Guided Adaptive Radiotherapy for Monitoring and Treating Glioblastomas” and “Is MRI Contrast Enhancement Necessary for Long-Term Monitoring of Diffuse Glioma?”)
For more insights from Dr. Kann, watch the video below.
Considering Breast- and Lesion-Level Assessments with Mammography AI: What New Research Reveals
June 27th 2025While there was a decline of AUC for mammography AI software from breast-level assessments to lesion-level evaluation, the authors of a new study, involving 1,200 women, found that AI offered over a seven percent higher AUC for lesion-level interpretation in comparison to unassisted expert readers.
Can CT-Based Deep Learning Bolster Prognostic Assessments of Ground-Glass Nodules?
June 19th 2025Emerging research shows that a multiple time-series deep learning model assessment of CT images provides 20 percent higher sensitivity than a delta radiomic model and 56 percent higher sensitivity than a clinical model for prognostic evaluation of ground-glass nodules.
FDA Clears Ultrasound AI Detection for Pleural Effusion and Consolidation
June 18th 2025The 14th FDA-cleared AI software embedded in the Exo Iris ultrasound device reportedly enables automated detection of key pulmonary findings that may facilitate detection of pneumonia and tuberculosis in seconds.