ACR Appropratiness Criteria (AC) NLP Search Tool

Repository to accompany manuscript

Purpose

Our study developed and evaluated a search and match natural language processing tool for selecting evidence-based radiology studies in the American College of Radiology (ACR) Appropriateness Criteria (AC) using patient clinical indications and demographics. The AC is underutilized by clinicians, resulting in less evidence-based care. We aim to increase use by developing an efficient, clinician-centered web app that can match clinical indications to AC documents and variants.

Instructions for Running Locally

1. pip install requirements.txt
2. Gather additional requirements (see below)
3. Run process_artificial_inds.ipynb jupyter notebook

Additional requirements

• Embeddings: our model has similar embeddings to publically availabe ones here.
• Word2int and Vocab files: These were too large to include in this repository, and can be substituted with those extracted from the above public model using a modified sent2vec implementation (/sent2vec-master-edited).

Acknowledgements

We are grateful to the developers of fasttext, sent2vec, and BioSentVec for making their software and dataset available publically.