Functional imaging constrained diffusion for brain PET synthesis from structural MRI.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Multimodal analysis of neurodegenerative disorders increasingly uses MRI and PET scans.
• PET scans are less accessible than MRI, limiting their widespread clinical use.
• Existing deep generative models for synthesizing PET from MRI often face training instability and fail to preserve functional brain information.

Purpose of the Study:

• To propose a novel Functional Imaging Constrained Diffusion (FICD) framework for synthesizing 3D brain PET images from structural MRI.
• To address limitations of existing methods by ensuring stable training and preserving functional information.
• To improve the fidelity and accuracy of synthesized PET scans for neurodegenerative disorder analysis.

Main Methods:

• Developed a Constrained Diffusion Model (CDM) within the FICD framework.
• Introduced a functional imaging constraint to the CDM to ensure voxel-wise alignment between synthesized and ground truth PET scans.
• Trained and evaluated the FICD framework on paired T1-weighted MRI and 18F-fluorodeoxyglucose (FDG)-PET scans from 293 subjects.

Main Results:

• FICD demonstrated superior performance in generating FDG-PET data compared to state-of-the-art methods in quantitative and qualitative analyses.
• The framework achieved high output fidelity through a stable training phase.
• Validated FICD's utility and generalizability on a larger dataset of 1262 subjects across three downstream tasks.

Conclusions:

• The proposed FICD framework offers a stable and effective method for synthesizing 3D brain PET images from MRI.
• FICD successfully preserves crucial functional brain information, outperforming existing methods.
• The framework shows significant potential for enhancing the analysis of neurodegenerative disorders and generalizability in medical imaging.