Peer-Reviewed Publication
Sci Rep2026;16(1)June 13, 2026Journal Article
Photoacoustic device fingerprints induce bias in deep learning models.
Christoph J Bender1,2, Marcel Knopp3,4, Niklas Holzwarth3,4, Tom Rix3,4, Jan-Hinrich Nölke3,4, Kris K Dreher3,5,6,7, Yi Li8, Julius Kempf8, Milenko Caranovic8, Fabian Schneider3,9,10, Melanie Schellenberg3,11, Leonie Boland3,4, Briain Haney8, Ferdinand Knieling12, Ulrich Rother8, Alexander Seitel3,13, Lena Maier-Hein14,15,16,17,18,19,20,21
1Division of Intelligent Medical Systems (IMSY), German Cancer Research Center (DKFZ) Heidelberg, Heidelberg, Germany. christophjulien.bender@dkfz-heidelberg.de.
2Medical Faculty, Heidelberg University, Heidelberg, Germany. christophjulien.bender@dkfz-heidelberg.de.
3Division of Intelligent Medical Systems (IMSY), German Cancer Research Center (DKFZ) Heidelberg, Heidelberg, Germany.
4Faculty of Mathematics and Computer Science, Heidelberg University, Heidelberg, Germany.
5Faculty of Physics and Astronomy, Heidelberg University, Heidelberg, Germany.
6Division of Medical Image Computing (MIC), German Cancer Research Center (DKFZ) Heidelberg, Heidelberg, Germany.
7Helmholtz Imaging, German Cancer Research Center (DKFZ), Heidelberg, Germany.
8Department of Vascular Surgery, University Hospital Erlangen, Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Erlangen, Germany.
9Institute of Biological and Medical Imaging, Helmholtz Zentrum München, Neuherberg, Germany.
10Chair of Biological Imaging, Central Institute for Translational Cancer Research (TranslaTUM), School of Medicine and Health, Technical University of Munich, Munich, Germany.
11Department of Medical Physics, Memorial Sloan Kettering Cancer Center, New York, USA.
12Department of Pediatrics and Adolescent Medicine, University Hospital Erlangen, Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Erlangen, Germany.
13National Center for Tumor Diseases (NCT), NCT Heidelberg, a partnership between DKFZ and University Hospital Heidelberg, Heidelberg, Germany.
14Division of Intelligent Medical Systems (IMSY), German Cancer Research Center (DKFZ) Heidelberg, Heidelberg, Germany. l.maier-hein@dkfz.de.
15Medical Faculty, Heidelberg University, Heidelberg, Germany. l.maier-hein@dkfz.de.
16Faculty of Mathematics and Computer Science, Heidelberg University, Heidelberg, Germany. l.maier-hein@dkfz.de.
17Helmholtz Imaging, German Cancer Research Center (DKFZ), Heidelberg, Germany. l.maier-hein@dkfz.de.
18National Center for Tumor Diseases (NCT), NCT Heidelberg, a partnership between DKFZ and University Hospital Heidelberg, Heidelberg, Germany. l.maier-hein@dkfz.de.
19Surgical AI Research Group, Heidelberg University Hospital, Surgical Clinic, Heidelberg, Germany. l.maier-hein@dkfz.de.
20HIDSS4Health - Helmholtz Information and Data Science School for Health, Karlsruhe/Heidelberg, Germany. l.maier-hein@dkfz.de.
21Mohamed Bin Zayed University of Artificial Intelligence (MBZUAI), Abu Dhabi, UAE. l.maier-hein@dkfz.de.
Abstract
Deep learning (DL) models developed for established medical imaging modalities have shown increasing performance and reliability as a result of scaling efforts. In contrast, model development for emerging modalities such as photoacoustic imaging (PAI) remains challenged by data sparsity, which limits model generalizability and raises the susceptibility to bias. While recent studies in PAI have sta…
Create a free account to keep reading
Free members get 10 full research views every month across publications, clinical trials, FDA clearances, adverse events, and NIH grants. No credit card required.
Want unlimited research access? See Pro plans