Generating Synthetic MR Perfusion Maps From DWI and FLAIR in Acute Ischemic Stroke: Development and External...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Magnetic resonance imaging (MRI) is crucial for acute stroke diagnosis but often requires time-consuming contrast-enhanced perfusion imaging.
• Current methods like dynamic susceptibility contrast (DSC) perfusion add complexity and delay treatment decisions.

Purpose of the Study:

• To develop a method for synthesizing T-map perfusion maps from noncontrast MRI sequences.
• To reduce reliance on contrast-enhanced perfusion imaging in acute stroke triage.
• To potentially shorten MRI protocols and accelerate stroke treatment.

Main Methods:

• A deep generative model, specifically a denoising diffusion probabilistic model with a 2.5D architecture, was trained.
• The model used diffusion-weighted imaging (DWI), fluid-attenuated inversion recovery (FLAIR), and infarct core masks as inputs.
• Performance was evaluated by comparing synthetic T-max maps against ground truth using image similarity metrics and Dice coefficients for T-max > 6s regions.

Main Results:

• The best-performing model generated synthetic T-max perfusion maps with high similarity to ground truth in under 110 seconds.
• Strong spatial overlap was observed for T-max > 6s regions in both internal (Dice: 0.82) and external (Dice: 0.59) validation.
• The infarct core mask, DWI, and FLAIR were critical inputs for accurate perfusion map synthesis.

Conclusions:

• A noninvasive, scalable framework was developed to generate synthetic T-max perfusion maps from noncontrast MRI.
• This approach can increase access to perfusion data in acute stroke settings.
• It has the potential to accelerate treatment decisions by eliminating the need for contrast-enhanced imaging.