Acceleration of EM-Based 3D CT Reconstruction Using FPGA.

Area of Science:

• Medical Imaging
• Computer Engineering
• Algorithm Optimization

Background:

• Reducing radiation dose in computed tomography (CT) is crucial for patient safety.
• Iterative algorithms like expectation maximization (EM) offer high-quality 3D reconstruction but suffer from long execution times.
• Current limitations hinder the clinical application of low-dose CT in time-sensitive scenarios.

Purpose of the Study:

• To significantly reduce the execution time of iterative CT reconstruction algorithms.
• To enable rapid low-dose 3D reconstruction for time-critical medical events.
• To develop an efficient hardware-accelerated solution for CT image processing.

Main Methods:

• A novel parallel processing scheme leveraging field-programmable gate arrays (FPGAs) and their block RAMs.
• An external memory bandwidth reduction strategy for efficient data reuse (sinogram and voxel intensity).
• A customized FPGA-based processing engine designed for high throughput and reduced logic consumption.
• A hardware and software co-design flow for adaptable CT machine integration.

Main Results:

• Achieved a 26.9x speedup compared to a 16-thread multicore CPU implementation.
• Demonstrated feasibility of completing reconstruction in minutes, suitable for time-critical applications.
• Successfully implemented the system on an FPGA-based server-class node using actual patient data.

Conclusions:

• The proposed FPGA-based parallel scheme dramatically accelerates low-dose CT 3D reconstruction.
• This advancement makes iterative reconstruction methods practical for time-critical medical imaging.
• The developed system offers a significant speedup, enhancing the clinical utility of low-dose CT.