HyperSLICE: HyperBand optimized spiral for low-latency interactive cardiac examination

Area of Science:

• Medical Imaging
• Cardiovascular MRI
• Deep Learning

Background:

• Interactive cardiac MRI is crucial for real-time applications like scan planning and MR-guided interventions.
• Current methods face limitations in spatio-temporal resolution due to real-time acquisition and visualization constraints.

Purpose of the Study:

• To enhance interactive cardiac MRI resolution by optimizing undersampled spiral sampling.
• To leverage deep learning for low-latency reconstruction, specifically deep artifact suppression.

Main Methods:

• A variable density spiral trajectory was optimized using HyperBand for rapid deep artifact suppression.
• The HyperSLICE framework was trained on 692 breath-held CINE datasets.
• Prospective testing in 13 subjects, including simulations and clinical interventions, compared HyperSLICE with conventional Cartesian real-time and breath-hold CINE imaging.

Main Results:

• Optimized spiral sampling significantly improved image quality metrics (RMS error, SNR, structural similarity, Laplacian energy) compared to radial and uniform spiral sampling in simulations.
• Prospectively, HyperSLICE achieved higher spatial and temporal resolution than conventional Cartesian real-time imaging.
• The system demonstrated rapid reconstruction suitable for interactive imaging during procedures like catheter pullback.

Conclusions:

• HyperSLICE enables high spatial and temporal resolution for interactive cardiac MRI.
• Optimized spiral sampling enhances overall image quality and transition handling compared to other trajectories.