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

2025-08-28 · Source: cs.CV updates on arXiv.org · Field: Science & Research — Health & Medical Research, Mathematics & Computational Sciences · Depth: Expert, extended

Summary

GLI-AL is a new controlled-access, labels-only resource derived from the BraTS 2023-GLI training cohort, designed to address label noise from unannotated white matter hyperintensities (WMH) in glioma MRI segmentation. It provides 1,251 unified eight-class anatomy-lesion label sets, aligned with original four-modal MRI cases, including image-repair labels for 116 cases. The resource categorizes cases into a 394-case purified subset and an 857-case extended subset, offering case-level metadata for label source, image-repair needs, and quality control. Compared to original BraTS-GLI annotations, GLI-AL significantly expands foreground supervision by incorporating healthy brain tissues and previously unlabeled coexisting abnormalities within a single label space. This addition accounts for 1.51 billion voxels, representing 92.7% of the released foreground, with 2.20 million additional lesion voxels in 857 cases, directly addressing the label-noise problem.

Key takeaway

For AI Scientists and Machine Learning Engineers working on medical image segmentation, GLI-AL offers a critical resource for improving model robustness. If you are developing models for glioma MRI, you should integrate this WMH-aware, unified anatomy-lesion label set to reduce label noise and enhance sensitivity to coexisting abnormalities. This allows for more accurate joint segmentation of healthy tissues and lesions, enabling better evaluation of data quality impacts on model performance.

Key insights

GLI-AL unifies healthy tissue and lesion labels in BraTS-GLI to mitigate white matter hyperintensity (WMH) label noise.

Principles

• Unlabeled abnormalities introduce task-specific label noise.
• Unified label spaces improve joint segmentation supervision.
• Data stratification by quality enables robust analysis.

Method

GLI-AL constructs a purified subset via expert-negative WMH cases and model screening, completes extended subset labels using DeepWMH/LST-AI intersection, then fuses TumorSynth probability maps with lesion constraints for unified 8-class labels.

In practice

• Use GLI-AL for joint healthy tissue and lesion segmentation.
• Analyze label noise sensitivity using purified vs. extended subsets.
• Stratify experiments by label source and quality control status.

Topics

• Glioma MRI Segmentation
• White Matter Hyperintensities
• Label Noise Analysis
• Multi-Modal MRI
• Medical Image Datasets
• Anatomy-Lesion Labels

Code references

• xyx200/brats-gli-anatomy-lesion-code

Best for: Computer Vision Engineer, AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• GLI-AL: A Multi-Modal Glioma MRI Label Resource with Unified Anatomy-Lesion Labels — GLI-AL unifies anatomy-lesion labels for glioma MRI, improving segmentation by including white matter hyperintensities.
• CA-GCL: Cross-Anatomy Global-Local Contrastive Learning for Robust 3D Medical Image Understanding — CA-GCL uses global-local contrastive learning and clinical-aware text augmentation to prevent representation collapse in 3D medical FVLP.
• Comparative evaluation of training strategies using partially labelled datasets for segmentation of white matter hyperintensities and stroke lesions in FLAIR MRI — Pseudolabeling is the most effective strategy for deep learning segmentation of confounding medical lesions using partially labeled MRI data.
• Taxlifier: Leveraging Disease Taxonomy for Enhanced Multi-Label Classification in Chest Radiography — Leveraging disease taxonomy in hierarchical multi-label classification significantly improves chest X-ray diagnosis accuracy and interpretability.
• HSQ-VLM: A Novel Spatially-Constrained Quadrant Segmentation VLM Model for Explainability in Diabetic Retinopathy — HSQ-VLM provides explainable DR diagnostics by segmenting fundus images with fovea-centered anatomical precision.
• Label-free multiphoton microscopy and machine learning for recognition of hepatocellular carcinoma — Label-free multiphoton microscopy combined with machine learning accurately identifies hepatocellular carcinoma.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.