Detectability and healthcare implications of generative AI–synthesized chest radiographs: a blinded radiologist reader study - Takeaways - MDSpire

Detectability and healthcare implications of generative AI–synthesized chest radiographs: a blinded radiologist reader study

  • By

  • Jinghang Wang

  • Ruixin Wang

  • Qijia Yi

  • Hongyong Tang

  • Li Fan

  • Shunan Lin

  • Wenjing He

  • Dan Peng

  • Jun-Jie Yang

  • Jun Liu

  • July 15, 2026

Share

  • 1

    The study evaluated synthetic chest radiographs generated by GPT-image and Gemini-image models using text-only and image-conditioned generation strategies.

  • 2

    A total of 320 real disease-positive chest radiographs were assessed alongside 320 matched normal conditioning radiographs in a blinded reader study.

  • 3

    Image-conditioned generation showed significantly lower AI detection rates compared to text-only generation, with rates of 34.0% versus 56.1%.

  • 4

    Synthetic radiographs from the GPT-image model were less frequently detected than those from the Gemini-image model, indicating varying performance.

  • 5

    The findings highlight the importance of transparent labeling and provenance tracking for synthetic medical images in healthcare and education.

Original Source(s)

Detectability and healthcare implications of generative AI–synthesized chest radiographs: a blinded radiologist reader study

  • by Jinghang Wang, Ruixin Wang, Qijia Yi, Hongyong Tang, Li Fan, Shunan Lin, Wenjing He, Dan Peng, Jun-Jie Yang, Jun Liu

  • July 15, 2026

Related Content

Longitudinal changes in the ALPS index and its clinical correlates in patients with basal ganglia hemorrhage

Beyond AUC: a clinician’s guide to building and trusting prediction models in oncology—a narrative review

Deep learning-based malignancy probability estimation of pulmonary nodules in PET/CT imaging