Coin Inbreast Eval

---
name: coin-inbreast-eval
description: Evaluates a deep learning model's ability to classify breast masses as benign or malignant in mammography images. It probes the effectiveness of adversarial data augmentation and contrastive manifold learning in improving discriminative feature extraction under data scarcity. Use when the user wants to benchmark on INbreast, or asks about evaluating this task. Reports Accuracy.
metadata:
  skill_kind: dataset_eval
  source_arxiv: 2012.14690
  bibtex_key: li2020coin
  confidence: high
---

# coin-inbreast-eval

> COIN: Contrastive Identifier Network for Breast Mass Diagnosis in Mammography — Li et al. (2020) (arXiv:2012.14690, 2020)

## What this evaluates

Evaluates a deep learning model's ability to classify breast masses as benign or malignant in mammography images. It probes the effectiveness of adversarial data augmentation and contrastive manifold learning in improving discriminative feature extraction under data scarcity.

## Datasets

- **INbreast** — total ?; splits: test (-1)

## Metrics

- `Accuracy` **(primary)** — range: percent
  - Percentage of correctly classified breast mass samples (benign vs. malignant) out of the total test set.
- `AUC` — range: [0, 1]
  - Area under the Receiver Operating Characteristic (ROC) curve, measuring the model's ability to distinguish between benign and malignant classes across all classification thresholds.

## Input / output format

**Input**: Regions of interest (ROIs) containing breast masses from FFDM mammography images, labeled as benign or malignant.

**Output**: Binary classification prediction (benign or malignant) and/or class probabilities for AUC computation.

## Scoring recipe

```python
def compute_metrics(predictions, labels):
    # predictions: list of predicted classes or probabilities
    # labels: list of ground truth classes (0=benign, 1=malignant)
    accuracy = sum(p == l for p, l in zip(predictions, labels)) / len(labels)
    auc = roc_auc_score(labels, predictions) # assuming probabilities or scores
    return accuracy, auc
```

## Common pitfalls

- State-of-the-art comparison results are taken directly from original papers rather than re-evaluated on the same split, making direct performance comparison potentially unfair.
- The dataset is small, and the reported metrics include standard deviations (e.g., 93.4 ± 1.9%), implying multiple runs or cross-validation, but the exact experimental protocol (e.g., number of folds or seeds) is not detailed.
- Visual quality assessment of generated augmentations is subjective and not quantitatively measured alongside the classification metrics.

## Evidence (verbatim from paper)

> To better evaluate the performance, we validate the proposed algorithm on the small FFDM mammography dataset: the INbreast dataset. ... From the Tab. [I], we can notice that all augmentation algorithms have improved the classification performance when comparing with the baseline model. ... COIN has outperformed the state-of-the-art with mean accuracy 93.4% and AUC score 0.95.

## Citation

```bibtex
@misc{li2020coin,
  title={COIN: Contrastive Identifier Network for Breast Mass Diagnosis in Mammography},
  author={Li et al. (2020)},
  year={2020},
  note={arXiv:2012.14690}
}
```

- arXiv: 2012.14690