Technical - Emerging CMR Methods - Acquisition & Reconstruction-Accelerated 2D Phase Contrast MRI using deep learning-based reconstruction and direct complex difference estimation

Accelerated 2D Phase Contrast MRI using deep learning-based reconstruction and direct complex difference estimation

Video Transcription

Video Summary

Julio Escanoa presented a deep learning framework for accelerated 2D phase-contrast MRI reconstruction. Building on DLE-SPIRIT, the team used phase-contrast and complex-difference networks to reconstruct undersampled velocity encodes, then estimate phase difference, velocity, peak velocity, and net flow. Using 194 fully sampled multi-coil datasets, they compared the method with compressed sensing (L1-ESPIRIT + TV). The deep learning approach outperformed compressed sensing and enabled up to 8x acceleration while keeping peak velocity and net flow errors within 5%, potentially reducing scan time or improving spatio-temporal resolution.

Keywords

deep learning

phase-contrast MRI

accelerated reconstruction

velocity encoding

compressed sensing