The abstract is read at the publisher; the summary is JClub's.
BraTS-Path 2024 benchmark for glioblastoma sub-regions shows top AI performance at Accuracy=0.69 and MCC=0.58, despite strong intra-tumoral heterogeneity and annotation variability.
- Why it matters: Automated classification of histopathologic sub-regions from whole slide images is needed for quantitative tumor characterization, but heterogeneity and inter-observer disagreement limit reliable model evaluation.
- What they did: Eight expert neuropathologists produced reference annotations for 309,118 H&E WSI samples across training/validation/hold-out splits of 96,485/62,387/150,246, and six AI approaches were ranked using multiple metrics including Accuracy and Matthew's correlation coefficient (MCC).
- The result: Top methods used large pre-trained models, ensembling, and extensive data augmentation, yet severe class imbalance and ambiguous labels like pseudopalisading necrosis (PN) and white matter penetration (WM) constrain minority-class performance.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.