Clinical Report: Assessing the Detectability and Clinical Impact of AI-Generated Chest Radiographs
Overview
This study evaluates the detectability of synthetic chest radiographs generated by AI models and their perceived visual authenticity by radiologists. Findings indicate that image-conditioned generation results in lower detection rates compared to text-only generation, highlighting the implications for clinical use and education.
Background
The integration of generative artificial intelligence (GenAI) in medical imaging presents both opportunities and challenges. As AI-generated images become more realistic, understanding their detectability is crucial to prevent contamination of clinical datasets and ensure proper educational use. This study addresses the gap in knowledge regarding the visual authenticity and clinical impact of synthetic chest radiographs.
Data Highlights
Generation Strategy
Detection Rate
Image-Conditioned Generation
34.0%
Text-Only Generation
56.1%
Key Findings
Image-conditioned generation had significantly lower AI detection rates than text-only generation (34.0% vs. 56.1%).
Synthetic radiographs from the GPT-image model were less frequently detected than those from the Gemini-image model.
Image-level structural and perceptual similarity was higher in GPT-image generated radiographs compared to Gemini-image.
Paired comparison improved detection of previously undetected image-conditioned synthetic radiographs.
There is a need for transparent labeling and expert review of synthetic medical images.
Clinical Implications
The findings suggest that synthetic chest radiographs may be perceived as authentic by radiologists, which raises concerns about their integration into clinical workflows. It is essential to implement measures for provenance tracking and expert review to mitigate potential risks associated with the use of synthetic images.
Conclusion
This study highlights the challenges posed by AI-generated chest radiographs in terms of detectability and authenticity. Ongoing evaluation and careful integration of these technologies into healthcare are necessary.
