Wienholt, Patrick https://orcid.org/0009-0007-8350-4288
Caselitz, Sophie
Siepmann, Robert
Bruners, Philipp
Bressem, Keno
Kuhl, Christiane
Kather, Jakob Nikolas
Nebelung, Sven
Truhn, Daniel
Funding for this research was provided by:
Deutsche Forschungsgemeinschaft (NE 2136/7-1, NE 2136/3-1, TR 1700/7-1)
German Federal Ministry of Research, Technology and Space BMFTR (Transform Liver - 031L0312C, DECIPHER-M - 01KD2420B)
German Federal Ministry of Research, Technology and Space BMFTR (PEARL, 01KD2104C; CAMINO, 01EO2101; TRANSFORM LIVER, 031L0312A; TANGERINE, 01KT2302 through ERA-NET Transcan; Come2Data, 16DKZ2044A; DEEP-HCC, 031L0315A; DECIPHER-M, 01KD2420A; NextBIG, 01ZU2402A)
European Union Research and Innovation Programme (ODELIA - GA 101057091)
German Cancer Aid DKH (DECADE, 70115166)
German Research Foundation DFG (CRC/TR 412, 535081457; SFB 1709/1 2025, 533056198)
German Academic Exchange Service DAAD (SECAI, 57616814)
German Federal Joint Committee G-BA (TransplantKI, 01VSF21048)
HORIZON EUROPE European Innovation Ecosystems (ODELIA, 101057091; GENIAL, 101096312)
European Research Council ERC (NADIR, 101114631)
National Institutes of Health NIH (EPICO, R01 CA263318)
National Institute for Health and Care Research NIHR (Leeds Biomedical Research Centre, NIHR203331)
Article History
Received: 3 November 2025
Revised: 17 December 2025
Accepted: 26 January 2026
First Online: 20 February 2026
Compliance with ethical standards
:
: The scientific guarantor of this publication is Daniel Truhn (dtruhn@ukaachen.de).
: The authors of this manuscript declare relationships with the following companies: D.T. received honoraria for lectures by Bayer, GE, Roche, AstraZeneca, and Philips, and holds shares in StratifAI GmbH, Germany, and in Synagen GmbH, Germany. J.N.K. declares consulting services for AstraZeneca, MultiplexDx, Panakeia, Mindpeak, Owkin, DoMore Diagnostics, and Bioptimus. Furthermore, he holds shares in StratifAI, Synagen, and Spira Labs, has received an institutional research grant from GSK, and has received honoraria from AstraZeneca, Bayer, Daiichi Sankyo, Eisai, Janssen, Merck, MSD, BMS, Roche, Pfizer, and Fresenius.
: One of the authors (D.T.) has significant statistical expertise and oversaw the study design and analyses.
: Written informed consent was not required for this study because it exclusively used publicly available, fully de-identified datasets and involved no interaction with human participants or access to identifiable personal data.
: Institutional Review Board approval was not required as the analysis used only publicly available, fully de-identified data with no patient contact and no access to identifiable personal information.
: This study reuses previously published, fully de-identified datasets with complete overlap and no new patient enrollment: VQA-Med 2019 (500 image-question pairs) and the diagnostic radiology dataset described by Huppertz et al (206 cases). The present work performs a new methodological evaluation of DSE for hallucination detection and selective answering in radiology VLMs; to the best of our knowledge, no prior publication has reported DSE on these datasets for this purpose.
: