Corebt Eval

---
name: corebt-eval
description: Evaluates multimodal fusion models for robust brain tumor typing by integrating MRI, histopathology, and diagnostic text under variable modality availability conditions. The benchmark probes a model's ability to perform fine-grained hierarchical classification across six glioma subtypes when some modalities are missing or degraded. Use when the user wants to benchmark on CoRe-BT, or asks about evaluating this task. Reports accuracy.
metadata:
  skill_kind: dataset_eval
  source_arxiv: 2603.03618
  bibtex_key: herasrivera2026corebt
  confidence: medium
---

# corebt-eval

> CoRe-BT: A Multimodal Radiology-Pathology-Text Benchmark for Robust Brain Tumor Typing — Heras Rivera et al. (2026) (arXiv:2603.03618, 2026)

## What this evaluates

Evaluates multimodal fusion models for robust brain tumor typing by integrating MRI, histopathology, and diagnostic text under variable modality availability conditions. The benchmark probes a model's ability to perform fine-grained hierarchical classification across six glioma subtypes when some modalities are missing or degraded.

## Datasets

- **CoRe-BT** — total 310; splits: train (-1), val (-1), test (-1)

## Metrics

- `accuracy` **(primary)** — range: [0, 1]
  - Standard classification accuracy measuring the proportion of correctly predicted glioma subtypes out of total patients. Exact metric variant (e.g., macro-F1) is not specified in the provided text.

## Input / output format

**Input**: Per patient: 3D MRI volumes, whole-slide histopathology images (processed into 256x256 tiles), and diagnostic text.

**Output**: Hierarchical tumor classification label across six glioma subtypes.

## Scoring recipe

```python
def compute_accuracy(predictions, gold_labels):
    correct = sum(1 for p, g in zip(predictions, gold_labels) if p == g)
    return correct / len(gold_labels)
```

## Common pitfalls

- Missing-modality robustness is a core evaluation condition, but specific missingness protocols, randomization seeds, or split configurations are not detailed in the provided text.
- Hierarchical labeling implies multi-level classification, but the exact evaluation hierarchy and how subtypes are mapped to labels are not specified.

## Evidence (verbatim from paper)

> The dataset includes 310 patients with expert-annotated tumor masks and pathologist-validated hierarchical labels across six glioma subtypes, enabling evaluation of multimodal fusion under missing-modality conditions. Baseline experiments demonstrate that multimodal models outperform MRI-only baselines, especially in distinguishing rare glioma subtypes.

## Citation

```bibtex
@misc{herasrivera2026corebt,
  title={CoRe-BT: A Multimodal Radiology-Pathology-Text Benchmark for Robust Brain Tumor Typing},
  author={Heras Rivera et al. (2026)},
  year={2026},
  note={arXiv:2603.03618}
}
```

- arXiv: 2603.03618