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
Clinical Scorecard: Assessing the Detectability and Clinical Impact of AI-Generated Chest Radiographs: A Study with Blinded Radiologist Readers
At a Glance
| Category | Detail |
|---|
| Condition | Synthetic Chest Radiographs |
| Key Mechanisms | Generative artificial intelligence (GenAI) for medical image synthesis |
| Target Population | Radiologists and medical imaging professionals |
| Care Setting | Medical imaging and healthcare education |
Key Highlights
- Evaluation of synthetic chest radiographs generated by GPT-image and Gemini-image models
- Comparison of text-only generation vs. image-conditioned generation
- Detection rates were significantly lower for image-conditioned synthetic radiographs
- Synthetic images may appear visually authentic to radiologists
- Need for transparent labeling and provenance tracking of synthetic images
Guideline-Based Recommendations
Diagnosis
- Systematic evaluation of visual realism and detectability of synthetic images is necessary.
Management
- Controlled integration of synthetic medical images into healthcare workflows is recommended.
Monitoring & Follow-up
- Expert review of synthetic images should be conducted to ensure quality and authenticity.
Risks
- Potential contamination of imaging datasets with synthetic radiographs.
Patient & Prescribing Data
Not applicable, as this study focuses on imaging technology rather than direct patient treatment.
Synthetic images may support dataset augmentation and medical education.
Clinical Best Practices
- Use report-informed prompts for generating synthetic images.
- Ensure transparency in labeling synthetic images.
- Implement provenance tracking for synthetic medical images.
Related Resources & Content
