Medical imaging and AI advancements improve glioma recurrence detection

New advancements in medical imaging and artificial intelligence are improving the ability to detect and differentiate glioma recurrence.

A temporal deep-learning model has been developed to predict one-year pediatric glioma recurrence using serial magnetic resonance imaging (MRI). By analyzing multiple historical scans, the model achieved an AUROC of 0.75 to 0.89, significantly improving F1 scores over standard longitudinal models. The study, which analyzed data from 715 pediatric patients across several datasets including the Dana-Farber Cancer Institute and the Children’s Brain Tumor Network, suggests that performance increases with the number of scans available, plateauing between three and six scans.

In parallel, targeted PET imaging using the amino acid tracer floretyrosine F 18 is showing promise in distinguishing recurrent malignancy from radiation necrosis. This is a critical clinical challenge, as conventional MRI often struggles to differentiate between tumor progression and post-treatment tissue changes like pseudoprogression. In a prospective multicenter study of 127 patients, the positive percent agreement for detecting true recurrence ranged between 70% and 83%, with an 83% agreement rate observed in pediatric patients.