README.md

---
pretty_name: MultiCaRe Case Images (representative)
license: cc-by-4.0
task_categories:
  - image-to-text
  - document-question-answering
language:
  - en
size_categories:
  - 10K<n<100K
---

# MultiCaRe: Open-Source Clinical Case Dataset

MultiCaRe is an open-source, multimodal clinical case dataset built from the PubMed Central Open Access (OA) Case Report articles. It aggregates de-identified, open-access case narratives, figure images, captions, and rich article metadata across diverse specialties. Figures are mapped to their case text with explicit references and to article-level metadata, enabling grounded multimodal use.

- Source and process: OA case reports were collected from PMC; article metadata and abstracts were parsed; figures were downloaded and split into subimages when needed; captions were aligned; and image labels were curated from a hierarchical medical taxonomy (>140 classes).
- Scale: 85k+ OA case reports, 160k+ images/subimages (v2.0).
- Tasks enabled: image-text retrieval, caption grounding, VQA/doc-QA, image classification, multimodal modeling.
- Citation: DATA journal paper — https://www.mdpi.com/2306-5729/10/8/123; Zenodo — https://zenodo.org/records/13936721.

This repository: one representative image per figure
One row per figure (image_id) with a representative processed image and textual context that mentions the figure in the case narrative.

Highlights
- Representative image chosen per figure (preference: undivided > 'a' > first available).
- Includes text_references snippets to ground the figure in the case text.
- Join to image-, case-, and article-level datasets using stable keys.

Schema
- image: datasets.Image (PIL-compatible)
- image_id: original figure identifier (groups subimages)
- file: processed image filename used as the representative
- caption: figure caption (original)
- text_references: newline-joined excerpts that mention this figure
- tag: PubMed file tag
- case_id: equals cases.case_id
- article_id: PMCID
- file_id, patient_id, license: cross-links and license string

Quick start
```python
from datasets import load_dataset
ds = load_dataset("openmed-community/multicare-case-images", split="train")

row = ds[0]
row["image"].show()
print(row["caption"])          # caption of the figure
print(row["text_references"])  # where it appears in the case text
```

Joins
- image_id ↔ images.main_image
- case_id ↔ cases.case_id
- article_id ↔ articles.article_id

Notes
- Prefer patient/article-level splits downstream.
- Per-item OA licenses are preserved.