robailleo/medical-imaging-combined

Combined Medical Imaging Dataset

Dataset Description

Combined medical imaging dataset with 6793 samples in Alpaca instruction format.

Dataset Statistics

• —Total Samples: 6793
• —Training Samples: 5434
• —Validation Samples: 1359

Modality Distribution

• —X-ray: 2691 samples
• —CT: 2257 samples
• —Unknown: 1329 samples
• —MRI: 369 samples
• —Ultrasound: 147 samples

Source Distribution

• —ROCO: 5000 samples
• —VQA-RAD: 1793 samples

Sources

1. 1.ROCO (Radiology Objects in Context): Medical images with detailed captions
2. 2.VQA-RAD: Visual question answering for radiology with expert annotations

Intended Use

• —Fine-tuning vision-language models on medical imaging tasks
• —Research in medical AI and healthcare applications
• —Educational purposes in medical imaging analysis

Format

Each sample follows the Alpaca instruction format:

{
  "instruction": "Medical task description",
  "input": "Additional context",
  "output": "Medical analysis or answer",
  "image": "PIL.Image",
  "dataset_source": "ROCO or VQA-RAD",
  "modality": "Imaging modality"
}

Ethical Considerations

• —For research purposes only
• —Not for clinical decision making
• —Comply with medical ethics guidelines

Citation

Please cite the original datasets:

• —ROCO: https://huggingface.co/datasets/eltorio/ROCOv2-radiology
• —VQA-RAD: https://huggingface.co/datasets/flaviagiammarino/vqa-rad

License

Apache 2.0