An Open Multilingual Radiology Reports Dataset

0 2 minutes read

[Submitted on 25 Jul 2025]

Authors:Bastieen Le Guellec, Kokou Adambonou, Lisa C Adams, Thibault Agripnidis, Sung Soo ahn, Rathia Ait Charal, Tugba Akinci D Antonoli, Philippe Amouyel, Henrik Andersson, Raphael Bentegec, Claudio Benzoni, Antonino Andrinino, Felix, Ricardo Cao, Armando Augo Cavalo, Christil Claudio Vanie, Carlos Ferrarotti, Claudia Fossaru, Folic, Michel Folic, Powell Jack, Martina Jashovska, Ignasio Garcia Khuraiz, Marco Gatti, Natalia Gorlik, Alexia Maria Juliano, Haglis Hamero, Nicholas Hernierina, Czezetv Krocik, Dominic Copeka, Quintin Hlayi, Philip. Raval Combanovsky, Alexander Levervry, Tristan Limki, Maximilian Lindhols, Lucas Muller, Pioter Massic, Marcus McCovsky, Luigi Manacio, Ameer Midib, Antonio Natalie, Patri Nujima Edzang, Adiana Oujid, Ponsiglione, Malgorzata Poreba, Rafal Poreba, Philipp Prucker, Jean Pierre Pruvo, Rosa Alba Pugliesi, Feno Hasina Rabmanorintso, Vasileios Rafailidis, Kaarzyna Rotkegel, LUCA SABA, Tekin, Lizin Toapanta Yanchapaxi, Matthaios Triantaillou, Ekaterini Tsaoulia, Evangelia Vassalou, Federica Vernuccio, Johan Wasselius, Weilang Wang, Sizymon Urban, Adrian Wlodarczak, Sizymon Wlodarck, Andrzej, Lina, LINA. Shthang Chang, Sebastian Zigmaleer, Gregory Kochsenski, Keno K Brigim

PDF view of the paper entitled Parrot: A group of open -language radiology reports, by Bastien Le Guellec and 86 other books

PDF HTML (experimental) view

a summary:The logical basis and objectives: to develop and verify the health of the parrot (the reports of the funny radiology for the open test), a large and multiple -centered data collection and an open space of fictional radiology reports that extend multiple languages to test natural language processing applications in the radiation. Materials and Roads: From May to September 2024, radiologists were invited to contribute to fictional radiology reports after their standard reports practices. The shareholders have submitted at least 20 reports with the descriptive data related to the anatomical area, the method of photography, the clinical context and non -English reports, English translations. All ICD-10 codes have been set. A study of the distinction of human report versus artificial intelligence was conducted with 154 participants (radiologists, healthcare professionals, and unhealthy care specialists) to assess whether the reports are composed of or created. Results: Data collection includes 2,658 radiology report of 76 books across 21 countries and 13 languages. Reports cover multiple photography methods (CT: 36.1 %, MRI: 22.8 %, radiography: 19.0 %, ultrasound: 16.8 %) and anatomical areas, with chest (19.9 %), abdomen (18.6 %), head (17.3 %), and bilif (14.1 %) is more appropriate. In the discrimination study, participants achieved a resolution of 53.9 % (95 % CI: 50.7 % -57.1 %) in distinguishing between human and AI reports, with a significantly better rays (56.9 %, 95 % CI: 53.3 % -60.6 %, P <0.05) of other groups. Conclusion: Parrot represents the largest collection of data on multi -language radiology reports, enabling the development of natural language processing applications and verification across the linguistic, geographical and clinical borders without privacy restrictions.