Can AI Make Paediatric PET/CT Lower Dose? Opportunities, Evidence and Cautions

Abstract

Artificial intelligence (AI) has emerged as a potential enabling technology for lower-dose paediatric PET/CT, but its value depends on whether dose reduction can be achieved without compromising diagnostic accuracy, lesion detectability, quantitative reliability or clinical confidence. This review examines AI-enabled pathways for paediatric PET/CT dose optimisation, including low-count PET denoising, image enhancement, full-count synthesis, reconstruction-integrated AI, low-dose CT reconstruction, synthetic CT, CT-free attenuation/scatter correction, motion correction, automated quality control, patient-specific protocol selection and reduction of unnecessary or repeat imaging. The strongest direct opportunities are low-count PET enhancement and AI-supported CT dose reduction, while synthetic CT and CT-free correction may reduce or eliminate attenuation-correction CT in selected settings where diagnostic CT is not required. Indirect AI approaches, including motion-risk reduction, quality-control systems and agentic workflow support, may reduce dose by improving first-time-right imaging, avoiding repeat acquisition and matching protocols to the clinical question. It is important to differentiate between visually aesthetic images and diagnostic images, in other words, visual image improvement is an inadequate endpoint. AI-enhanced lower-dose PET/CT must be validated for small-lesion detection, quantitative accuracy, response-classification stability, scanner and tracer generalisability, failure modes and prospective clinical safety. Social, ethical and legal considerations are also central, because AI tools have the potential to widen disparities. AI may make paediatric PET/CT lower dose, but only when it preserves clinical value within a governed, paediatric-specific optimisation framework.

Artificial intelligence (AI) has emerged as a potential enabling technology for lower-dose paediatric PET/CT, but its value depends on whether dose reduction can be achieved without compromising diagnostic accuracy, lesion detectability, quantitative reliability or clinical confidence. This review examines AI-enabled pathways for paediatric PET/CT dose optimisation, including low-count PET denoising, image enhancement, full-count synthesis, reconstruction-integrated AI, low-dose CT reconstruction, synthetic CT, CT-free attenuation/scatter correction, motion correction, automated quality control, patient-specific protocol selection and reduction of unnecessary or repeat imaging. The strongest direct opportunities are low-count PET enhancement and AI-supported CT dose reduction, while synthetic CT and CT-free correction may reduce or eliminate attenuation-correction CT in selected settings where diagnostic CT is not required. Indirect AI approaches, including motion-risk reduction, quality-control systems and agentic workflow support, may reduce dose by improving first-time-right imaging, avoiding repeat acquisition and matching protocols to the clinical question. It is important to differentiate between visually aesthetic images and diagnostic images, in other words, visual image improvement is an inadequate endpoint. AI-enhanced lower-dose PET/CT must be validated for small-lesion detection, quantitative accuracy, response-classification stability, scanner and tracer generalisability, failure modes and prospective clinical safety. Social, ethical and legal considerations are also central, because AI tools have the potential to widen disparities. AI may make paediatric PET/CT lower dose, but only when it preserves clinical value within a governed, paediatric-specific optimisation framework.