GLI-AL: A Multi-Modal Glioma MRI Label Resource with Unified Anatomy-Lesion Labels

Researchers introduced GLI-AL, a controlled-access glioma MRI label resource for joint anatomy-lesion segmentation.

It provides unified 8-class annotations, purified and extended subsets, and repair labels to account for coexisting white matter hyperintensities in BraTS-GLI.

By explicitly labeling these otherwise unlabeled findings, GLI-AL reduces label noise and supports more reliable training and evaluation.

In external tests, a MedNeXt model trained on the purified subset improved lesion sensitivity and reduced boundary error versus a noisy-control model.