About this Event
Recovering Brain PET at Reduced Dose: Generative Models, Physics Constraints, and
Quantitative Fidelity
Abstract: Brain Positron Emission Tomography (PET) measures molecular processes that structural imaging cannot capture, but its clinical and research utility is shaped by an inescapable trade-off between injected radioactivity and image quality. Lowering dose reduces radiation exposure, which is a priority for pediatric, longitudinal, and repeated-imaging settings, yet it also amplifies noise, blurs subtle regional contrast, and destabilizes the quantitative measures on which diagnosis and kinetic analysis depend. This talk presents research that reframes low-dose brain PET as a coupled signal recovery and quantification problem rather than a generic denoising task. The talk will trace the progression from SMART-PET, a self-similarity-aware generative adversarial framework that demonstrated diagnostic-quality recovery of [¹⁸F]FDG brain PET at 90% dose reduction across epilepsy, frontotemporal dementia, and healthy cohorts, to a current physics-guided generative framework that unifies low-count synthesis and dose recovery within a single scanner-aware architecture. Findings indicate that pairing patient-conditioned diffusion models with projector-based sinogram-space consistency improves quantitative fidelity and cross-scanner stability over image-domain methods alone, and that dual-pathway synthesis enables training on archives that lack raw sinogram data. Together, this work charts a path toward clinically usable low-dose brain PET that preserves not just appearance but measurement.
Bio: Coming soon
Unable to attend in person? Join us virtually
For more information, visit The Neuro's website .
Event Venue & Nearby Stays
Montreal Neurological Institute – Hospital, 3801 Rue University, Montréal, Canada
USD 0.00
