Enhancing Amyloid PET Quantification: MRI-Guided Super-Resolution Using Latent Diffusion Models.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Physics

Background:

• Amyloid PET imaging is vital for Alzheimer's disease (AD) diagnosis and research.
• Low PET spatial resolution causes partial volume effects (PVE), limiting accurate amyloid quantification.
• Developing methods to overcome PVE is crucial for precise AD assessment.

Purpose of the Study:

• To introduce a novel latent diffusion model for resolution recovery (LDM-RR) to address PVE in amyloid PET imaging.
• To enhance the accuracy of amyloid deposition quantification in AD.
• To improve the detection of longitudinal changes and reduce variability in PET measurements.

Main Methods:

• A synthetic data generation pipeline was used to create high-resolution PET digital phantoms for training.
• The LDM-RR model employed a weighted combination of L1, L2, and MS-SSIM losses for MRI-guided reconstruction.
• Model performance was evaluated for improving statistical power in longitudinal change detection and inter-tracer agreement.

Main Results:

• The LDM-RR approach significantly improved PET quantification accuracy.
• Inter-tracer variability in amyloid PET measurements was reduced.
• The model enhanced the detection of subtle, time-dependent changes in amyloid deposition.

Conclusions:

• Deep learning, specifically the LDM-RR model, holds significant potential for improving PET quantification in AD research.
• This method can contribute to earlier and more accurate detection and monitoring of Alzheimer's disease progression.
• LDM-RR offers a promising tool for advancing AD diagnosis and therapeutic evaluation.