Breast Mri Lr Seg Eval

---
name: breast-mri-lr-seg-eval
description: Evaluates the ability of 3D medical image segmentation models to accurately delineate left and right breast tissues in MRI scans. It probes anatomical partitioning robustness and generalization across diverse clinical sources. Use when the user wants to benchmark on Breast MRI Left-Right Segmentation Dataset, or asks about evaluating this task. Reports Dice Similarity Coefficient (DSC).
metadata:
  skill_kind: dataset_eval
  source_arxiv: 2507.13830
  bibtex_key: rokuss2025divid
  confidence: high
---

# breast-mri-lr-seg-eval

> Divide and Conquer: A Large-Scale Dataset and Model for Left-Right Breast MRI Segmentation — Rokuss et al. (2025) (arXiv:2507.13830, 2025)

## What this evaluates

Evaluates the ability of 3D medical image segmentation models to accurately delineate left and right breast tissues in MRI scans. It probes anatomical partitioning robustness and generalization across diverse clinical sources.

## Datasets

- **Breast MRI Left-Right Segmentation Dataset** — total 13752; splits: train (-1), val (-1), test (-1)

## Metrics

- `Dice Similarity Coefficient (DSC)` **(primary)** — range: [0, 1]
  - Measures the overlap between predicted segmentation and ground truth, calculated as 2 * |A ∩ B| / (|A| + |B|), where A and B are the predicted and ground truth masks respectively.

## Input / output format

**Input**: 3D breast MRI volumes (volumetric image data)

**Output**: Binary segmentation masks delineating the left and right breast regions

## Scoring recipe

```python
def compute_dsc(pred_mask, gt_mask):
    intersection = np.sum(pred_mask & gt_mask)
    union = np.sum(pred_mask) + np.sum(gt_mask)
    if union == 0:
        return 1.0
    return 2.0 * intersection / union
# Average DSC across all folds and patients
```

## Common pitfalls

- The dataset combines volumes from diverse clinical sources, so performance may vary significantly across scanners or protocols not seen during training.
- Cross-validation is performed at the subject level, but the exact splitting strategy to prevent patient data leakage is not detailed in the results section.

## Evidence (verbatim from paper)

> To assess the performance and generalizability of our nnU-Net model for left and right breast segmentation in MRI scans, we conducted a 5-fold cross-validation. The dataset was partitioned into five subsets, ensuring that each fold served as a validation set once while the remaining four were used for training. The segmentation accuracy was evaluated using the Dice Similarity Coefficient (DSC), a standard metric for measuring the overlap between the predicted segmentation and the ground truth.

## Citation

```bibtex
@misc{rokuss2025divid,
  title={Divide and Conquer: A Large-Scale Dataset and Model for Left-Right Breast MRI Segmentation},
  author={Rokuss et al. (2025)},
  year={2025},
  note={arXiv:2507.13830}
}
```

- arXiv: 2507.13830