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

[Submitted on 29 Jul 2025]

View the PDF file for the paper entitled Rexgroundingct: 3D Data collection CT to divide the results from free text reports, by Muhammad Bahrun and 22 other authors

PDF HTML (experimental) view

a summary:We offer Rexgroundingct, the first data collection available to the public to link the results of the radiology in the free text with sectors at the pixel level in 3D tests in the manually explained CT scan. While previous data groups relied on organized posters or pre -defined categories, Rexgroundingct captures the full expression of the clinical language represented in the free text and the reasons for 3D retailers translated in volumetric photography. This deals with a critical gap in medical artificial intelligence: the ability to connect the complex prescription text, such as “3 mm nodules in the left lower lobe”, to its precise anatomical location in the three -dimensional space, which is an essential ability to generate infinite radiology report. The data collection includes 3,142 uninterrupted chest tests associated with unified radiology reports from the CT rate data collection. Using a three-phase pipeline, GPT-4 was used to extract positive results of the lung and fond, which was manually divided by expert broadcasters. A total of 8,028 results were explained by 16,301 entities, with the quality monitoring of the radiologists approved by the Board of Directors. Nearly 79 % of the results are pivotal distortions, while 21 % are not interested. The training group includes up to three representative sectors per discovery, while health and test verification groups contain comprehensive stickers for each entity. Rexgroundingct creates a new standard for developing and evaluating basic models at the sentence level and the free -text retail models in CT CT. The data collection can be accessed in this URL https.