Can we really trust generative models in healthcare? A systematic review of uncertainty quantification in generative AI for medical imaging.
Comput Methods Programs BiomedJul 21, 2026 (epub)
Clinical ResearchRadiology (Diagnostic)
Alen Shahini, Silvia Seoni, Alessandra Manzin et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review evaluated the integration of uncertainty quantification (UQ) methods in generative AI for medical imaging, analyzing 41 studies published from 2018 to 2025. The review found that UQ integration was often linked to improved performance metrics, such as reconstruction quality and segmentation accuracy, with 56% of studies published in 2025, indicating rapid growth in the field.
Key findings
- 29 of the 41 studies focused on radiology, highlighting its prominence in the research.
- 56% of the studies were published in 2025, indicating rapid growth in the field.
- UQ integration was frequently associated with improved performance in reconstruction quality and segmentation accuracy.
- Key priorities include standardizing uncertainty metrics and developing efficient computational frameworks.
Why it matters
Understanding the role of uncertainty quantification in generative AI can enhance the reliability of medical imaging applications, which is crucial for clinical decision-making and patient safety.