Deep learning denoising enables accurate SUV measurement in half-count PET scans

AI denoising software PETfectior achieved 99.9% lesion detection sensitivity on half-count PET/CT scans with only one false positive. SUVmax bias was -1.01% (95% limits of agreement ±12.5%), and image quality was not significantly different from full-count scans.

• 1649 lesions were analyzed across 198 FDG-PET/CT studies in 258 patients.
• One false positive was detected among the half-count + AI images.
• Subjective image quality was equivalent between half-count + AI and standard full-count images.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.