Synapse
⌘+K
Synapse
PulseExploreJournal ClubResearchersJournals
Instagram
HomeJournal ClubExplore
June 28, 2021Open Access

RadGraph: Extracting Clinical Entities and Relations from Radiology Reports

View Full Paper
Ask AI
Bookmark
Share

Authors

SJSaahil JainAAAshwin AgrawalASAdriel Saporta

Discussion

Loading...

Member takes

Overview

Randomized trial demonstrates efficient extraction of clinical information from radiology reports, suggesting new applications in healthcare.

Key Points

  • The research focuses on developing RadGraph for extracting structured information from radiology reports to improve healthcare applications.
  • Developed a novel information extraction schema for radiology reports.
  • Created a development dataset of 500 radiology reports with board-certified radiologist annotations.
  • Trained a deep learning model, RadGraph Benchmark, for evaluating relation extraction accuracy.
  • Achieved a micro F1 score of 0.82 on the MIMIC-CXR test set for relation extraction.
  • Achieved a micro F1 score of 0.73 on the CheXpert test set for relation extraction.
  • Generated an inference dataset with annotations across 220,763 MIMIC-CXR reports, enhancing data availability.

Cite This Study

Jain et al. (2021) studied this question.

synapsesocial.com/papers/69d755fff182769aa8b8a5c0https://doi.org/10.48550/arxiv.2106.14463
View Full Paper
Ask AI
Bookmark
Share