Glioblastoma cells spread beyond tumour boundaries visible on MRI, making radiotherapy targeting challenging. Researchers from DQBM’s Menze Lab, the Technical University of Munich and international partners developed PREDICT-GBM to systematically evaluate computational approaches to this problem.
The platform combines longitudinal imaging from 243 patients across three datasets with a standardised evaluation pipeline. In retrospective comparisons, leading deep-learning and biophysical models achieved modest improvements in coverage of regions where tumours later recurred, without increasing the target volume.
By making data, models and evaluation tools openly available, PREDICT-GBM enables reproducible comparisons and further method development. Whether these improvements translate into better patient outcomes requires clinical validation.
Read the publication in npj Digital Medicine