Breast Cancer Mammogram Eval

---
name: breast-cancer-mammogram-eval
description: Evaluates deep learning models for breast cancer lesion detection and classification in mammogram images, comparing segmentation and classification performance against conventional and heuristic-optimized baselines. Use when the user wants to benchmark on Dataset 1 & 2, or asks about evaluating this task. Reports accuracy.
metadata:
  skill_kind: dataset_eval
  source_arxiv: 2310.19411
  bibtex_key: [REDACTED]
  confidence: high
---

# breast-cancer-mammogram-eval

> Intelligent Breast Cancer Diagnosis with Heuristic-assisted Trans-Res-U-Net and Multiscale DenseNet using Mammogram Images — Yaqub et al. (2023) (arXiv:2310.19411, 2023)

## What this evaluates

Evaluates deep learning models for breast cancer lesion detection and classification in mammogram images, comparing segmentation and classification performance against conventional and heuristic-optimized baselines.

## Datasets

- **Dataset 1 & 2** — total ?; splits: test (-1)

## Metrics

- `accuracy` **(primary)** — range: percent
  - Proportion of correctly classified instances out of the total number of instances. Calculated as (TP + TN) / (TP + TN + FP + FN).
- `precision` — range: percent
  - Proportion of true positive predictions among all positive predictions. Calculated as TP / (TP + FP).
- `f1-score` — range: percent
  - Harmonic mean of precision and recall (sensitivity). Calculated as 2 * TP / (2 * TP + FP + FN).
- `sensitivity` — range: percent
  - Proportion of actual positives correctly identified. Calculated as TP / (TP + FN).
- `specificity` — range: percent
  - Proportion of actual negatives correctly identified. Calculated as TN / (TN + FP).
- `mcc` — range: [-1, 1]
  - Matthews Correlation Coefficient measures the quality of binary classifications. Calculated as (TP*TN - FP*FN) / sqrt((TP+FP)(TP+FN)(TN+FP)(TN+FN)).
- `jaccard-distance` — range: [0, 1]
  - Intersection over Union (IoU) between the predicted segmentation mask and the ground truth mask. Calculated as |Prediction ∩ Ground Truth| / |Prediction ∪ Ground Truth|.

## Input / output format

**Input**: Mammogram images with corresponding ground truth masks and binary labels for segmentation and classification tasks.

**Output**: Predicted segmentation masks and classification labels (binary) for each mammogram image.

## Scoring recipe

```python
def compute_metrics(tp, tn, fp, fn):
    accuracy = (tp + tn) / (tp + tn + fp + fn)
    precision = tp / (tp + fp) if (tp + fp) > 0 else 0
    f1 = 2 * tp / (2 * tp + fp + fn) if (2 * tp + fp + fn) > 0 else 0
    sensitivity = tp / (tp + fn) if (tp + fn) > 0 else 0
    specificity = tn / (tn + fp) if (tn + fp) > 0 else 0
    denom = ((tp+fp)*(tp+fn)*(tn+fp)*(tn+fn))**0.5
    mcc = (tp*tn - fp*fn) / denom if denom > 0 else 0
    return accuracy, precision, f1, sensitivity, specificity, mcc
```

## Common pitfalls

- Dataset sizes and train/val/test splits are not explicitly stated in the text.
- Metrics are reported as percentages in tables, but the formulas use raw counts (TP, TN, FP, FN).
- Jaccard distance is computed on segmentation masks, while the other metrics are classification-based.

## Evidence (verbatim from paper)

> The below-provided metrics are utilized in assessing the implemented BC detection framework. Negative Predictive Value (NPV) Np is determined by Eq. (12). The precision pcn is evaluated based on Eq. (13). False Discovery Rate (FDR) Fd is computed as in Eq. (14). Specificity Sp is determined as in Eq. (15). Matthews Correlation Co-efficient (MCC) Mc is evaluated as provided in Eq. (16). Sensitivity Se is calculated using Eq. (17). The F1 score Fs is determined using Eq. (18). False Negative Rate (FNR) Fn is evaluated using Eq. (19). The Jaccard distance Jd between the ground truth image/mask images and the segmented image is computed using Eq. (20).

## Citation

```bibtex
@misc{yaqub2023intelligent,
  title={Intelligent Breast Cancer Diagnosis with Heuristic-assisted Trans-Res-U-Net and Multiscale DenseNet using Mammogram Images},
  author={Yaqub et al. (2023)},
  year={2023},
  note={arXiv:2310.19411}
}
```

- arXiv: 2310.19411