Edge Computing for Physics-Driven AI in Computational MRI: A Feasibility Study.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Edge Computing

Background:

• Physics-driven AI (PD-AI) accelerates MRI scans, enabling higher resolutions.
• High-resolution MRI generates massive data, straining transmission, storage, and processing, especially in functional MRI.
• Edge computing with FPGAs offers solutions for near-sensor PD-AI reconstruction, but requires hardware-efficient models.

Purpose of the Study:

• To propose a novel PD-AI computational MRI approach optimized for FPGA-based edge computing.
• To enhance hardware efficiency through 8-bit complex data quantization and elimination of FFT/IFFT operations.

Main Methods:

• Developed a PD-AI computational MRI approach tailored for FPGA edge devices.
• Implemented 8-bit complex data quantization for model optimization.
• Eliminated redundant Fast Fourier Transform (FFT) and Inverse FFT (IFFT) operations.

Main Results:

• Achieved improved computational efficiency compared to conventional PD-AI methods.
• Maintained reconstruction quality comparable to existing PD-AI techniques.
• Outperformed standard clinical MRI methods in reconstruction quality and efficiency.

Conclusions:

• The proposed PD-AI approach enables high-resolution MRI reconstruction on resource-constrained edge devices.
• This strategy addresses data bottlenecks in high-resolution MRI, facilitating real-world deployment.
• Optimized PD-AI models are crucial for efficient edge computing in advanced medical imaging.