Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
June 21, 2026NEJM AIOpen Access

ReXGroundingCT: A 3D Chest CT Dataset for Segmentation of Findings from Free-Text Reports

View Full Paper
Ask AI
Bookmark
Share

Authors

MBMohammed BaharoonLLLuyang LuoMMMichael Moritz

Discussion

Loading...

Member takes

Overview

Randomized trial demonstrates accurate 3D segmentation from free-text CT findings, highlighting potential in medical AI.

Key Points

  • The aim is to create a dataset linking free-text findings to precise 3D segmentations in chest CT scans.
  • Introduced ReXGroundingCT, a dataset with 3142 noncontrast CT scans linked to standardized reports.
  • Used GPT-4 to extract and standardize findings, followed by categorization into a hierarchical ontology.
  • Produced 3D annotations validated by board-certified radiologists, with an additional chain-of-thought dataset.
  • Offers 16,301 annotated entities across 8028 text-to-3D segmentation pairs, covering various findings.
  • 79% of findings are focal abnormalities and 21% are nonfocal.
  • A public validation set of 50 cases and a private test set of 100 cases are available for evaluation.

Cite This Study

Baharoon et al. (2026) studied this question.

synapsesocial.com/papers/6a377f3a24f042ddf4c59f02https://doi.org/10.1056/aidbp2501220
View Full Paper
Ask AI
Bookmark
Share