CT-free attenuation correction of ¹³N-ammonia cardiac PET images using conditional denoising diffusion implicit model...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nuclear Medicine

Background:

• CT-based attenuation correction (CT-AC) in cardiac PET increases radiation exposure, posing risks for pediatric patients and those requiring repeat scans.
• Developing CT-free attenuation correction methods is crucial for reducing radiation dose in cardiac PET imaging.

Purpose of the Study:

• To develop and evaluate a novel conditional denoising diffusion implicit model (cDDIM) for generating CT-free attenuation-corrected cardiac PET images.
• To assess the performance of cDDIM against traditional generative adversarial network (GAN)-based methods using quantitative metrics.

Main Methods:

• A 2.5D cDDIM framework was employed, utilizing non-attenuation-corrected (NAC) PET input to synthesize CT-AC images.
• Two cDDIM versions were tested: cDDIM with logarithmic linear normalization (cDDIM_LLN) and cDDIM with linear normalization (cDDIM_LN).
• Performance was evaluated using normalized mean square error (NMSE) and segment-wise absolute percent error (APE) on data from two centers.

Main Results:

• Both cDDIM_LLN and cDDIM_LN visually outperformed GAN-based methods in generating attenuation-corrected cardiac PET images.
• cDDIM_LLN demonstrated significantly lower NMSE and APE compared to cDDIM_LN and GAN-based methods across both centers.
• cDDIM_LN also showed improved APE over GAN-based methods.

Conclusions:

• cDDIM-based methods successfully synthesized cardiac PET tracer distributions comparable to clinical CT-AC.
• The cDDIM_LLN approach yielded the best performance among the evaluated methods.
• cDDIM-based attenuation correction represents a significant advancement over traditional GAN-based techniques, offering a promising CT-free solution.