Published first-author IEEE TRPMS paper
🚀 I’ve got some early-access news from my PhD: a new first-authored paper is now live in IEEE Transactions on Radiation and Plasma Medical Sciences! 🎉
Title: Personalized MR-Informed Diffusion Models for 3D PET Image Reconstruction 🔍 What’s the idea? When someone has a PET brain scan, we often also have their MRI. Instead of training one “generic” AI model for everyone, we build a personalized prior:
- We warp PET–MR data from many subjects onto the person’s own MRI to create subject-specific “pseudo-PET” examples that match their anatomy.
- We pre-train a diffusion model on these personalized examples, so the model “expects” the right anatomy for that individual before we ever reconstruct their PET.
- During reconstruction, this MR-informed prior nudges the result toward realistic anatomy while still letting PET-unique uptake patterns show through (i.e. it avoids just copying MR).
- On low-count [18F]FDG data (both simulated and real), this personalized approach improves reconstruction accuracy versus non-personalized diffusion models, which is especially helpful when dose (and noise) are limiting factors.
đź‘Ą Big thanks to Alexander Hammers, Andrew King & Andrew Reader for their guidance and collaboration.
đź“„ Read it here:
- IEEE Early Access: https://ieeexplore.ieee.org/document/11138025
- Author’s accepted manuscript (arXiv): https://arxiv.org/abs/2506.03804
- Bibliographic details