Published first-author IEEE TMI paper

🚀 Excited to share a big milestone in my PhD: my first-author paper has just been accepted by IEEE Transactions on Medical Imaging! 🎉

Title: Likelihood-Scheduled Score-Based Generative Modeling for Fully 3D PET Image Reconstruction 🔍 What’s inside?

• Marrying physics & generative AI – we sync each denoising step of a pre-trained diffusion model with the classical ML-EM update.
• Fast reconstruction – we eliminate unnecessary data fidelity updates by taking dynamic likelihood updates.
• Easier model calibration – just one tunable hyperparameter sets how much the reconstruction leans on the learned diffusion prior versus the measured PET data.
• Fully 3-D diffusion-based PET reconstructions of real clinical data – we stack perpendicular 2D diffusion models to eliminate slice inconsistency issues.

Huge thanks to my co-authors, without whom this work wouldn’t have been possible!

📄 Read it here:

• IEEE Early-access
• Author’s accepted manuscript
• Bibliographic details

---