Accelerating GRAPPA reconstruction using SoC design for real-time cardiac MRI.

Area of Science:

• Medical Imaging
• Hardware Acceleration
• Cardiovascular Diseases

Background:

• Real-time cardiac MRI (CMR) is crucial for diagnosing cardiovascular diseases but faces challenges with high frame rates and temporal resolution.
• Existing methods like compressed sensing and parallel MRI (e.g., GRAPPA) improve CMR quality but demand substantial computational power, limiting real-time applications.
• Field-programmable gate arrays (FPGAs) offer a potential hardware solution to accelerate computationally intensive MRI reconstruction algorithms.

Purpose of the Study:

• To develop and evaluate a novel 32-bit floating-point FPGA-based GRAPPA accelerator.
• To enhance the speed and efficiency of cardiac MR image reconstruction for real-time clinical applications.
• To explore the trade-offs between reconstruction time, resource utilization, and design effort using FPGAs.

Main Methods:

• Designed a custom FPGA architecture featuring dedicated computational engines (DCEs) for continuous data flow between GRAPPA calibration and synthesis stages.
• Integrated a high-speed DDR4-SDRAM module for multi-coil MR data storage and an on-chip ARM Cortex-A53 processor for data transfer management.
• Implemented the accelerator on a Xilinx Zynq UltraScale+ MPSoC using high-level synthesis (HLS) and hardware descriptive language (HDL).
• Evaluated performance using in-vivo cardiac datasets (18- and 30-receiver coils) and compared with CPU and GPU-based GRAPPA methods.

Main Results:

• The FPGA-based GRAPPA accelerator achieved speed-up factors of up to 121x compared to CPU-based methods and 9x compared to GPU-based methods.
• Demonstrated reconstruction rates of up to approximately 27 frames-per-second.
• Maintained the visual quality of reconstructed cardiac MR images.

Conclusions:

• The proposed FPGA-based GRAPPA accelerator significantly accelerates cardiac MR image reconstruction.
• This hardware solution enables higher frame rates and real-time imaging capabilities for clinical applications.
• FPGA acceleration offers a promising approach to overcome computational bottlenecks in advanced MRI techniques.