Breast Cancer Classification Eval

breast-cancer-classification-eval

Attend what matters: Leveraging vision foundational models for breast cancer classification using mammograms — Sanghvi et al. (2026) (arXiv:2604.19350, 2026)

What this evaluates

Evaluates a vision model's ability to classify mammograms as benign or malignant, focusing on fine-grained discrimination of localized malignancies and handling high-resolution medical images.

Datasets

• Public mammogram dataset(s) — total ?; splits: (unstated)

Metrics

• F1 score (primary) — range: [0, 1]
  • Harmonic mean of precision and recall for the binary classification task. Computed without averaging across classes.
• AUC — range: [0, 1]
  • Area under the Receiver Operating Characteristic curve, measuring the model's ability to distinguish between classes across all classification thresholds.

Input / output format

Input: High-resolution mammogram images (processed at 1024x1024 or via extracted Regions of Interest).

Output: Binary classification label (malignant vs. non-malignant).

Scoring recipe

def compute_metrics(preds, labels):
    tp = sum(p == 1 and l == 1 for p, l in zip(preds, labels))
    fp = sum(p == 1 and l == 0 for p, l in zip(preds, labels))
    fn = sum(p == 0 and l == 1 for p, l in zip(preds, labels))
    prec = tp / (tp + fp) if (tp + fp) > 0 else 0
    rec = tp / (tp + fn) if (tp + fn) > 0 else 0
    f1 = 2 * prec * rec / (prec + rec) if (prec + rec) > 0 else 0
    auc = roc_auc_score(labels, preds)
    return {'F1': f1, 'AUC': auc}

Common pitfalls

• Metrics are computed strictly under a binary classification setting without averaging across classes.
• Baseline comparisons may be biased if multimodal models are evaluated with empty text strings for non-malignant cases due to missing findings in the reference dataset.
• Reproducing prior SOTA is difficult as the previous best model (Mammo-CLIP) was trained on a private dataset.

Evidence (verbatim from paper)

All scores here are computed under the binary classification setting, without averaging across classes. ... our scores demonstrate a 4% improvement on the F1 score and a 1% gain on the AUC over this SOTA

Citation

@misc{sanghvi2026attend,
  title={Attend what matters: Leveraging vision foundational models for breast cancer classification using mammograms},
  author={Sanghvi et al. (2026)},
  year={2026},
  note={arXiv:2604.19350}
}

• arXiv: 2604.19350