E2D-unroll: efficient equivariant deformable unrolling networks for cardiac cine MRI reconstruction

Key Points

• The aim is to enhance cardiac cine MRI reconstruction by reducing artifacts while maintaining efficiency.
• Developed the E2D-Unroll network combining deformable convolutions with rotation equivariance.
• Evaluated performance in terms of aliasing reduction and computational efficiency.
• Implemented in real-world cardiac cine MRI settings.
• Significantly suppressed large-scale aliasing artifacts compared to traditional methods.
• Maintained a high parameter efficiency with low computational cost.
• Demonstrated effectiveness in accelerated imaging scenarios.

Abstract