ACR Appropriateness Criteria Corpus

Dataset Card for the ACR Appropriateness Criteria Corpus

This dataset contains chunked guidelines and narratives from the ACR Appropriateness Criteria, an set of societal guidelines from the American College of Radiology (ACR) to help clinicians order appropriate diagnostic imaging studies for patients. The corpus is formatted similarly to the corpuses introduced in MedRAG by Xiong et al. (2024), and can therefore be similarly used for medical Retrieval-Augmented Generation (RAG).

Please abide by the ACR Terms and Conditions when using this dataset.

Dataset Structure

Each row of the dataset contains the following fields:

{
  "id": "A unique identifier for each chunk in the corpus",
  "title": "The title of the main document from which the chunk was extracted",
  "content": "The content of the chunk",
  "contents": "The title and content concatenated together into a single field",
  "ACRID": "The ACR topic identifier"
}

Of note, the contents field for each row does not exceed a maximum of 2048 characters in our implementation.

Dataset Curation

This dataset was curated by scraping the ACR Appropriateness Criteria website, downloading all the relevant PDF narratives, and using unstructured's open-source tooling to extract the content into chunks. The original source code used for the dataset curation can be found here. This dataset was constructed on July 17, 2024 - no updates to the ACR Appropriateness Criteria after this date are reflected in the most recent iteration.

Personal and Sensitive Information

The contents of this dataset consist of evidence-based guidelines written by a panel of experts for a medical audience. To our knowledge, no personal or sensitive information is contained in this dataset.

Dataset Card Contact

Please contact Michael Yao at michael.yao@pennmedicine.upenn.edu with any questions or comments.