Subcellsam Eval

---
name: subcellsam-eval
description: Evaluates zero-shot segmentation accuracy on cellular and subcellular microscopy imagery, and assesses the downstream reliability of extracted morphological features for drug hit validation in high-content screening assays. Use when the user wants to benchmark on Cell segmentation datasets, Hit validation datasets, or asks about evaluating this task. Reports Dice Score (DSC), Z'-factor.
metadata:
  skill_kind: dataset_eval
  source_arxiv: 2508.13701
  bibtex_key: hanimann2025subcellsam
  confidence: high
---

# subcellsam-eval

> subCellSAM: Zero-Shot (Sub-)Cellular Segmentation for Hit Validation in Drug Discovery — Hanimann et al. (2025) (arXiv:2508.13701, 2025)

## What this evaluates

Evaluates zero-shot segmentation accuracy on cellular and subcellular microscopy imagery, and assesses the downstream reliability of extracted morphological features for drug hit validation in high-content screening assays.

## Datasets

- **Cell segmentation datasets** — total ?; splits: test (-1)
- **Hit validation datasets** — total ?; splits: test (-1)

## Metrics

- `Dice Score (DSC)` **(primary)** — range: [0, 1]
  - Computes twice the intersection between predicted and ground truth masks divided by the sum of their areas. Standard convention for binary segmentation overlap.
- `Intersection over Union (IoU)` — range: [0, 1]
  - Calculates the ratio of the intersection area to the union area between predicted and ground truth masks. Standard Jaccard index for segmentation.
- `Z'-factor` **(primary)** — range: other
  - Assays quality metric derived from positive and negative control distributions: 1 - 3*(σ_p + σ_n) / |μ_p - μ_n|. Used to evaluate assay robustness in hit validation.
- `EC50` — range: other
  - Half-maximal effective concentration, calculated from dose-response curves fitted to extracted morphological/intensity features. Indicates compound potency.

## Input / output format

**Input**: Raw microscopy images containing cellular, nuclear, and subcellular structures.

**Output**: Binary segmentation masks at cell, nucleus, and subcellular entity levels.

## Scoring recipe

```python
def compute_segmentation_metrics(pred_mask, gt_mask):
    intersection = np.logical_and(pred_mask, gt_mask).sum()
    dsc = 2 * intersection / (gt_mask.sum() + pred_mask.sum())
    iou = intersection / np.logical_or(pred_mask, gt_mask).sum()
    return dsc, iou

# Downstream:
# 1. Extract morphological/intensity features from binary masks.
# 2. Fit dose-response curves to feature values across compound titrations.
# 3. Compute Z'-factor from control variances/means.
# 4. Extract EC50 as the concentration yielding 50% of max response.
```

## Common pitfalls

- Comparing zero-shot inference against fine-tuned baselines without controlling for training data distribution or architectural differences.
- Assuming fixed hyperparameters generalize optimally across diverse microscopy modalities without dataset-specific tuning.
- Confusing segmentation quality metrics (DSC/IoU) with downstream assay quality/potency metrics (Z'-factor/EC50).

## Evidence (verbatim from paper)

> To assess the performance and usability of subCellSAM for cell segmentation we employ two different metrics: the Dice Score (DSC) and the Intersection over Union (IoU). ... To assess the performance and usability of subCellSAM for Downstream analysis we employ two different metrics: Z’-factor and EC50 which are the predominant result read-outs in the biopharma industry for hit validation use cases.

## Citation

```bibtex
@misc{hanimann2025subcellsam,
  title={subCellSAM: Zero-Shot (Sub-)Cellular Segmentation for Hit Validation in Drug Discovery},
  author={Hanimann et al. (2025)},
  year={2025},
  note={arXiv:2508.13701}
}
```

- arXiv: 2508.13701