{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,7,3]],"date-time":"2026-07-03T06:19:21Z","timestamp":1783059561423,"version":"3.54.6"},"reference-count":29,"publisher":"Frontiers Media SA","license":[{"start":{"date-parts":[[2026,7,3]],"date-time":"2026-07-03T00:00:00Z","timestamp":1783036800000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"DOI":"10.13039\/501100001659","name":"Deutsche Forschungsgemeinschaft","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100001659","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":["frontiersin.org"],"crossmark-restriction":true},"short-container-title":["Front. Radiol."],"abstract":"<jats:sec>\n                    <jats:title>Background<\/jats:title>\n                    <jats:p>\n                      The choice of image reconstruction approach, such as Iterative Model Reconstruction (IMR) and iDose\n                      <jats:inline-formula>\n                        <mml:math xmlns:mml=\"http:\/\/www.w3.org\/1998\/Math\/MathML\" id=\"IM1\">\n                          <mml:msup>\n                            <mml:mi\/>\n                            <mml:mn>4<\/mml:mn>\n                          <\/mml:msup>\n                        <\/mml:math>\n                      <\/jats:inline-formula>\n                      , is critical in computed tomography (CT) as distribution shifts between them can introduce systematic bias in clinical AI models, posing a barrier to efficient, scalable AI deployment in radiology. While Image-to-Image (I2I) translation offers a path toward harmonization, efficient and translation-feasible AI in radiology demands that such approaches be validated not only on image quality metrics but also on downstream clinical task performance.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Methods<\/jats:title>\n                    <jats:p>We compared three harmonization strategies: CycleGAN, Denoising Diffusion Models, and a conventional Gaussian filter, using a paired CT dataset. Performance was evaluated using task-agnostic metrics, including Structural Similarity Index Measure (SSIM), Peak Signal-to-Noise Ratio (PSNR), Frechet Inception Distance (FID), and Sliced Wasserstein Distance (SWD). Crucially, we assessed clinical utility by measuring the downstream impact on multi-organ segmentation using Mean Relative Volume Error (RVE), focusing on consistency across reconstructed modes.<\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Results<\/jats:title>\n                    <jats:p>\n                      Our findings reveal a significant discrepancy between visual fidelity and clinical utility. While Diffusion and CycleGAN models achieve competitive feature-level similarity, they introduce catastrophic systematic biases in downstream segmentation, particularly in adipose tissues (Body Composition Analysis, BCA), where volume errors can reach\n                      <jats:inline-formula>\n                        <mml:math xmlns:mml=\"http:\/\/www.w3.org\/1998\/Math\/MathML\" id=\"IM2\">\n                          <mml:mo>\u2212<\/mml:mo>\n                        <\/mml:math>\n                      <\/jats:inline-formula>\n                      100%, indicating a failure to preserve critical anatomical fat structures. In contrast, the computationally efficient Gaussian filter served as the most robust baseline, maintaining minimal RVE and the highest consistency between IMR and iDose\n                      <jats:inline-formula>\n                        <mml:math xmlns:mml=\"http:\/\/www.w3.org\/1998\/Math\/MathML\" id=\"IM3\">\n                          <mml:msup>\n                            <mml:mi\/>\n                            <mml:mn>4<\/mml:mn>\n                          <\/mml:msup>\n                        <\/mml:math>\n                      <\/jats:inline-formula>\n                      modes.\n                    <\/jats:p>\n                  <\/jats:sec>\n                  <jats:sec>\n                    <jats:title>Conclusions<\/jats:title>\n                    <jats:p>The results suggest that in the context of efficient AI for radiology imaging, where efficiency encompasses not only computational cost but also reliability of clinical outputs, task-agnostic image quality metrics are insufficient proxies for downstream clinical performance. High-parameter generative models may introduce structural losses that remain invisible to such metrics yet critically affect clinical utility. The underperformance of generative approaches observed here is likely attributable to 3D slice inconsistencies during volumetric reconstruction and suboptimal model adaptation, rather than to an inherent limitation of I2I methods, which retain significant potential for mitigating reconstruction-related bias in radiology AI. Future work developing harmonization pipelines must therefore prioritize task-specific, clinically grounded validation as the primary benchmark\u2014and carefully consider whether computationally simpler approaches, such as conventional filtering, may already provide comparably stable harmonization for specific use cases, before resorting to high-parameter generative solutions.<\/jats:p>\n                  <\/jats:sec>","DOI":"10.3389\/fradi.2026.1875867","type":"journal-article","created":{"date-parts":[[2026,7,3]],"date-time":"2026-07-03T05:40:44Z","timestamp":1783057244000},"update-policy":"https:\/\/doi.org\/10.3389\/crossmark-policy","source":"Crossref","is-referenced-by-count":0,"title":["Potential of Image2Image translation in reducing AI bias attributed to differences in CT reconstruction methods: proof-of-concept study on a paired dataset"],"prefix":"10.3389","volume":"6","author":[{"given":"Feifei","family":"Li","sequence":"first","affiliation":[{"name":"Institute for Biomedical Informatics, University of Cologne, Faculty of Medicine and University Hospital Cologne","place":["Cologne, Germany"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Liliana Lourenco","family":"Caldeira","sequence":"additional","affiliation":[{"name":"Institute for Diagnostic and Interventional Radiology, University of Cologne, Faculty of Medicine and University Hospital Cologne","place":["Cologne, Germany"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Astha","family":"Jaiswal","sequence":"additional","affiliation":[{"name":"Institute for Diagnostic and Interventional Radiology, University of Cologne, Faculty of Medicine and University Hospital Cologne","place":["Cologne, Germany"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Nils Gro\u00dfe","family":"Hokamp","sequence":"additional","affiliation":[{"name":"Institute for Diagnostic and Interventional Radiology, University of Cologne, Faculty of Medicine and University Hospital Cologne","place":["Cologne, Germany"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Oya","family":"Beyan","sequence":"additional","affiliation":[{"name":"Institute for Biomedical Informatics, University of Cologne, Faculty of Medicine and University Hospital Cologne","place":["Cologne, Germany"]},{"name":"FAIR Data and Distributed Analytics Department, Fraunhofer Institute for Applied Information Technology, FIT","place":["Aachen, Germany"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ekaterina","family":"Kutafina","sequence":"additional","affiliation":[{"name":"Institute for Biomedical Informatics, University of Cologne, Faculty of Medicine and University Hospital Cologne","place":["Cologne, Germany"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1965","published-online":{"date-parts":[[2026,7,3]]},"reference":[{"key":"B1","doi-asserted-by":"publisher","first-page":"e210290","DOI":"10.1148\/ryai.210290","article-title":"Mitigating bias in radiology machine learning: 1. Data handling","volume":"4","author":"Rouzrokh","year":"2022","journal-title":"Radiol Artif Intell"},{"key":"B2","doi-asserted-by":"publisher","first-page":"e220010","DOI":"10.1148\/ryai.220010","article-title":"Mitigating bias in radiology machine learning: 2. Model development","volume":"4","author":"Zhang","year":"2022","journal-title":"Radiol Artif Intell"},{"key":"B3","doi-asserted-by":"publisher","first-page":"e220061","DOI":"10.1148\/ryai.220061","article-title":"Mitigating bias in radiology machine learning: 3. Performance metrics","volume":"4","author":"Faghani","year":"2022","journal-title":"Radiol Artif Intell"},{"key":"B4","doi-asserted-by":"publisher","first-page":"e220171","DOI":"10.1148\/ryai.220171","article-title":"Hitting the mark: reducing bias in ai systems","volume":"4","author":"Kahn","year":"2022","journal-title":"Radiol Artif Intell"},{"key":"B5","doi-asserted-by":"publisher","first-page":"161","DOI":"10.1007\/s00330-018-5810-7","article-title":"The evolution of image reconstruction for CT\u2014from filtered back projection to artificial intelligence","volume":"29","author":"Willemink","year":"2018","journal-title":"Eur Radiol"},{"key":"B6","doi-asserted-by":"publisher","first-page":"109526","DOI":"10.1016\/j.ejrad.2021.109526","article-title":"Effect of CT reconstruction settings on the performance of a deep learning based lung nodule cad system","volume":"136","author":"Blazis","year":"2021","journal-title":"Eur J Radiol"},{"key":"B7","author":"Kl\u00e4ser","year":""},{"key":"B8","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1109\/JBHI.2021.3103387","article-title":"Acoustic simulation for transcranial focused ultrasound using GAN-based synthetic CT","author":"Koh","year":"2021","journal-title":"IEEE J Biomed Health Inform"},{"key":"B9","doi-asserted-by":"publisher","first-page":"101320","DOI":"10.1016\/j.patter.2025.101320","article-title":"A consensus privacy metrics framework for synthetic data","volume":"6","author":"Pilgram","year":"2025","journal-title":"Patterns"},{"key":"B10","author":"Isola","year":""},{"key":"B11","doi-asserted-by":"publisher","first-page":"3853","DOI":"10.1038\/s41598-025-87358-0","article-title":"Similarity and quality metrics for mr image-to-image translation","volume":"15","author":"Dohmen","year":"2025","journal-title":"Sci Rep"},{"key":"B12","author":"Krishnan","year":""},{"key":"B13","doi-asserted-by":"publisher","first-page":"451","DOI":"10.3390\/s22218540","article-title":"Unsupervised image-to-image translation: a review","volume":"22","author":"Hoyez","year":"2022","journal-title":"Sensors"},{"key":"B14","author":"Zhu","year":""},{"key":"B15","author":"Ho","year":""},{"key":"B16","doi-asserted-by":"publisher","first-page":"e230024","DOI":"10.1148\/ryai.230024","article-title":"Totalsegmentator: robust segmentation of 104 anatomic structures in ct images","volume":"5","author":"Wasserthal","year":"2023","journal-title":"Radiol Artif Intell"},{"key":"B17","doi-asserted-by":"publisher","first-page":"8750648","DOI":"10.1155\/2022\/8750648","article-title":"Deep learning-based image reconstruction for different medical imaging modalities","volume":"2022","author":"Yaqub","year":"2022","journal-title":"Comput Math Methods Med"},{"key":"B18","doi-asserted-by":"publisher","first-page":"321","DOI":"10.1038\/s41597-024-03073-x","article-title":"Brain tumor segmentation using synthetic mr images - a comparison of GANs and diffusion models","volume":"11","author":"Akbar","year":"2024","journal-title":"Sci Data"},{"key":"B19","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1145\/3643034","article-title":"Voxel-wise medical image generalization for eliminating distribution shift","volume":"18","author":"Li","year":"","journal-title":"ACM Trans Knowl Discov Data"},{"key":"B20","author":"Zhu","year":""},{"key":"B21","author":"Liu","year":""},{"key":"B22","volume-title":"Digital Image Processing Global Edition","author":"Gonzalez","year":"2017"},{"key":"B23","doi-asserted-by":"publisher","first-page":"600","DOI":"10.1109\/TIP.2003.819861","article-title":"Image quality assessment: from error visibility to structural similarity","volume":"13","author":"Wang","year":"2004","journal-title":"IEEE Trans Image Process"},{"key":"B24","author":"Heusel","year":""},{"key":"B25","author":"Szegedy","year":""},{"key":"B26","author":"Karras","year":""},{"key":"B27","year":""},{"key":"B28","doi-asserted-by":"publisher","first-page":"241","DOI":"10.1148\/radiol.211561","article-title":"Value-added opportunistic ct screening: state of the art","volume":"303","author":"Pickhardt","year":"2022","journal-title":"Radiology"},{"key":"B29","doi-asserted-by":"publisher","first-page":"726","DOI":"10.3233\/SHTI240517","article-title":"How data infrastructure deals with bias problems in medical imaging","volume":"316","author":"Li","year":"2024","journal-title":"Stud Health Technol Inform"}],"container-title":["Frontiers in Radiology"],"original-title":[],"link":[{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fradi.2026.1875867\/full","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,7,3]],"date-time":"2026-07-03T05:40:44Z","timestamp":1783057244000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.frontiersin.org\/articles\/10.3389\/fradi.2026.1875867\/full"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,7,3]]},"references-count":29,"alternative-id":["10.3389\/fradi.2026.1875867"],"URL":"https:\/\/doi.org\/10.3389\/fradi.2026.1875867","relation":{},"ISSN":["2673-8740"],"issn-type":[{"value":"2673-8740","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,7,3]]},"article-number":"1875867"}}