A Spatio-Temporal Diffusion Model for Cardiac Real-Time Imaging.

Area of Science:

• Medical Imaging
• Cardiovascular MRI
• Image Reconstruction

Background:

• Real-time cardiac imaging offers shorter scan times and is crucial for patients with arrhythmias or breath-holding difficulties.
• Achieving high spatio-temporal resolution in accelerated, undersampled cardiac MRI necessitates advanced reconstruction techniques.

Purpose of the Study:

• To investigate image quality improvements using a novel spatio-temporal diffusion-based reconstruction for accelerated spiral real-time cardiac MRI.
• To compare the proposed method against existing 2D spatial diffusion and compressed sensing techniques.

Main Methods:

• Clinical study utilizing accelerated spiral sampling for real-time cardiac acquisition during breath-hold and free-breathing.
• Training a spatio-temporal diffusion model using retrospectively binned segmented spiral cine images.
• Reconstruction and quantitative/qualitative assessment of accelerated acquisitions using the proposed model and baseline methods.

Main Results:

• Real-time acquisitions demonstrated shorter scan durations and improved quality, especially for participants with irregular heartbeats.
• Quantitative metrics and expert reader scores indicated superior image quality, sharpness, and reduced noise with the proposed spatio-temporal diffusion model.
• The method showed enhanced consistency between frames in reconstructions of undersampled spiral acquisitions.

Conclusions:

• Incorporating temporal information into the diffusion model enhances consistency, reduces noise, and preserves sharpness in undersampled spiral cardiac MRI.
• While promising, long reconstruction times and high computational demands remain challenges for clinical implementation.