README.md

---
license: cc-by-nc-4.0
task_categories:
- image-classification
language:
- en
tags:
- microscopy
- biomedical-imaging
- image-classification
- histology
- cerebellum
- mouse-brain
- lurcher
pretty_name: Lurcher 10x Microscopy
configs:
- config_name: images
  drop_labels: true
  data_files:
  - split: train
    path:
    - images/**/*.jpg
    - metadata.csv
---

# Lurcher 10x Microscopy Dataset

## Dataset overview

This dataset consists of 2-D microscopy images of histologically stained 3-D structures in tissue sections through the cerebellum of 21 mouse brains. Animals are grouped into wild-type controls (n = 10) and Lurcher mutant mice (n = 11). The classification task is to distinguish Lurcher mutant mice from wild-type controls.

All images were captured at low magnification (10x) and stained with Cresyl violet, a general histological stain for brain cells. Cresyl violet stains neurons, glia, endothelial cells, and other biostructures containing Nissl substance. The 10x magnification provides a balance between sufficient resolution to observe relevant cellular structures and a broad view of tissue architecture.

This dataset has also been used in Active Prompt Tuning (APT) experiments. Images are organized once by class and case id, while the experiment folds are provided as JSONL manifests plus fold-specific seed prompts.

## Acknowledgements

We thank Dr. Jan Cendelín and Dr. Yaroslav Kolinko at the Faculty of Medicine in Pilsen, Charles University, Czech Republic, for collecting and providing the tissue images and associated metadata. We are also grateful for their support in making this dataset publicly available so that the broader biomedical imaging and machine learning communities can use it.

## License and attribution

This dataset is released under the [Creative Commons Attribution-NonCommercial 4.0 International License](https://creativecommons.org/licenses/by-nc/4.0/). You may share and adapt the dataset for non-commercial purposes with appropriate attribution. Commercial use requires separate written permission from the rights holders.

If you use the dataset, fold manifests, seed prompts, or prompt bank in research, cite the associated APT paper listed below and retain attribution to the dataset contributors.

## Image layout

```text
images/microscopy_lurcher/
  lurcher/<case_id>/<image_name>.jpg
  wild/<case_id>/<image_name>.jpg
```

## Fold manifests

Predefined folds are included under:

```text
folds/microscopy_lurcher/fold-<fold>/
  train.jsonl
  val.jsonl
  test.jsonl
  seed.json
```

Each row has a path relative to the dataset repository root and a normalized class label:

```json
{"image_path":"images/microscopy_lurcher/lurcher/5917/example.jpg","class":"lurcher"}
```

`train.jsonl` is the active-learning candidate pool. `val.jsonl` is the validation split used by the APT code. `test.jsonl` is the held-out evaluation split. `seed.json` contains the initial seed prompt examples used before active selection starts.

## Dataset Viewer

The Hub dataset viewer is configured as an image dataset using:

```text
images/**/*.jpg
metadata.csv
```

`metadata.csv` links each image to its relative path, class, and case id so the viewer can render thumbnails directly while preserving the original `image_path` column. The predefined experiment folds remain in the JSONL manifests below `folds/`.

## Metadata

- `metadata.csv`: one row per image for the Dataset Viewer, with `file_name`, `image_path`, class, and case id.
- `metadata/images.csv`: one row per image with class and case id.
- `metadata/folds.csv`: per-fold split counts and manifest file paths.

## Prompt bank

The Lurcher prompt bank used by the APT oracle is included at:

```text
prompt_banks/microscopy_lurcher.json
```

Download it into the APT code repository with:

```bash
hf download USF-CS-Microscopy-Image-Analysis/Lurcher_10x \
  prompt_banks/microscopy_lurcher.json \
  --repo-type dataset \
  --local-dir .
```

The APT repository also includes `setup_lurcher_data.py`, which downloads the images, fold manifests, seed prompts, metadata, and prompt bank into the expected local layout automatically.

## Intended use

Use these fixed folds when comparing against the APT paper or repository results. Do not regenerate splits for direct comparison to reported fold-level numbers.

## Citation

```bibtex
@inproceedings{kandiyana2026apt,
  title={A Human-in-the-Loop Framework for Efficient Prompt Selection in Microscopy Vision-Language Models},
  author={Kandiyana, Abhiram and Mali, Ankur and Hall, Lawrence O. and Mouton, Peter R. and Goldgof, Dmitry},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
  pages={6229-6238},
  year={2026}
}
```