# Idh Mutation Prediction Eval

> Evaluates a model's ability to predict IDH mutation status (mutant vs. wild-type) in glioma patients using multi-modal MRI-derived structural brain networks. Use when the user wants to benchmark on TCIA + In-house Cohort, or asks about evaluating this task. Reports accuracy.

- Skill: `qhjqhj00/idh-mutation-prediction-eval` (Agent Skill)
- Install (CLI): `npx skillmds add qhjqhj00/idh-mutation-prediction-eval`
- Raw SKILL.md: https://api.skillmd.com/api/skills/qhjqhj00/idh-mutation-prediction-eval/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Data & Analytics
- Author: qhjqhj00 (https://skillmd.com/u/qhjqhj00)
- Updated: 2026-09-08
- Page: https://skillmd.com/skills/qhjqhj00/idh-mutation-prediction-eval

---


# idh-mutation-prediction-eval

> Predicting isocitrate dehydrogenase mutation status in glioma using structural brain networks and graph neural networks — Wei et al. (2021) (arXiv:2109.01854, 2021)

## What this evaluates

Evaluates a model's ability to predict IDH mutation status (mutant vs. wild-type) in glioma patients using multi-modal MRI-derived structural brain networks.

## Datasets

- **TCIA + In-house Cohort** — total 372; splits: train (-1), test (-1)

## Metrics

- `accuracy` **(primary)** — range: [0, 1]
  - Standard binary classification accuracy; exact evaluation metrics are not explicitly listed in the provided text, but binary cross-entropy is used for training.

## Input / output format

**Input**: Graph representation of a patient's structural brain network (90 nodes, 2309 edges) with 12-dimensional latent features per node/edge derived from multi-modal MRI (T1, T1c, T2, FLAIR).

**Output**: Binary label: IDH mutant or IDH wild-type.

## Scoring recipe

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

## Common pitfalls

- Autoencoder features are extracted using a model trained on the same 80% training split, potentially introducing data leakage or overfitting to the training cohort.
- Class imbalance (103 mutant vs 269 wild-type) is addressed via weighted loss rather than resampling, which may bias the model toward the majority class if weights are not carefully tuned.

## Evidence (verbatim from paper)

> This study included the pre-operative multi-modal MRI (pre-contrast T1, post-contrast T1, T2, and FLAIR) of 389 glioma patients. MRI images of 274 patients were downloaded from The Cancer Imaging Archive (TCIA) website, whereas 115 patients were available from an in-house cohort. 17 of 389 patients who have missing IDH mutation status or incomplete MRI modalities were excluded. For the included patients, 103 patients are IDH mutant, and 269 are IDH wild-type.

## Citation

```bibtex
@misc{wei2021idhmutation,
  title={Predicting isocitrate dehydrogenase mutation status in glioma using structural brain networks and graph neural networks},
  author={Wei et al. (2021)},
  year={2021},
  note={arXiv:2109.01854}
}
```

- arXiv: 2109.01854