ESPIRiT-Diffusion: Physics-Guided Diffusion Model Reconstruction for Highly Accelerated Joint Intracranial and Carotid Vessel Wall Imaging

Proposes ESPIRiT-Diffusion, embedding multi-set ESPIRiT coil sensitivity map-based data-consistency constraints into the Langevin equation to form a physics-guided score-based diffusion reconstruction framework. Relative to existing unfolding-based methods and k-space diffusion methods, this framework applies multi-set ESPIRiT map constraints directly within the diffusion sampling process to constrain unreliable generation at high acceleration. Supported by the methodological description and retrospective experiments (Cartesian, CAIPI, and variable-density undersampling at 8.8× and 10.7×) with comparisons against ESPIRiT, DL-ESPIRiT, SENSE-Diffusion, and SPIRiT-Diffusion.

Performing diffusion directly in the image domain yields clearer fine details and faster reconstruction. Compared with k-space diffusion, image-domain diffusion alleviates image blurring and reduces reconstruction time. Based on the text's comparison between image-domain and k-space diffusion, representing method design and experimental observation.

Achieves 0.6 mm³ isotropic resolution joint intracranial and carotid VWI reconstruction at 8.8× and 10.7× acceleration with better preservation of fine vessel wall structures. Extends highly accelerated diffusion reconstruction to large-FOV 3D joint intracranial and carotid VWI rather than a single vascular territory. Retrospective experiments showed improved performance over four comparison methods and better preservation of fine vessel wall structures.

In prospective patient experiments, ESPIRiT-Diffusion provided favorable visualization of vessel wall lesions, with no statistically significant differences in reader scores from the 3-fold CS reference across individual vascular segments or Overall comparisons. Moves retrospective reconstruction advantages into a prospective patient setting, with reader scores compared against a clinical reference sequence. Reader score comparisons in prospective patient experiments, with no statistically significant differences reported.

This work targets highly accelerated 3D joint intracranial and carotid VWI settings, applicable to MRI workflows with multi-coil acquisition and ESPIRiT sensitivity map estimation; its retrospective validation covers Cartesian, CAIPI, and variable-density undersampling, and prospective validation compares reader scores against a 3-fold CS reference. The results support reconstructing vessel wall details at 8.8× and 10.7× acceleration with 0.6 mm³ isotropic resolution, providing a basis for subsequent evaluation in broader patient populations and additional acquisition protocols.

The currently loaded text is summary-level and does not include specific sample sizes, reader scoring scales, statistical test details, image quality metric values, or the number of lesions in the prospective experiments, so effect sizes and statistical power cannot be assessed; additionally, performance differences across acceleration factors and sampling patterns, as well as reproducibility in other vascular territories or scanners, still require confirmation in the full methods and results.