FYP Part A: MRI Reconstruction on FPGA Hardware - Interim Report

Other sections of the interim report may be put into their respective posts as individual research pieces.

As part B of my FYP commences, I will be logging the progress and challenges that arise in a separate post.

---

Abstract

This project outlines a solution to the increasing computational demand of Magnetic Resonance (MR) image reconstruction, and the limitations of general-purpose Central Processing Units (CPUs) and Graphics Processing Units (GPUs) in meeting the low latency demands of clinical imaging.

The conducted literature review establishes cartesian k-space sampling and the Fast Fourier Transform (FFT) as the most appropriate reconstruction framework for a proof-of-concept hardware implementation. This FFT approach, coupled with Finite Impulse Response (FIR) low-pass filtering in the k-space domain, is the preferred pre-processing approach to demonstrate the versatility of specialised hardware.

The target platform is the Terasic Cyclone V GX Starter Kit, featuring the Cyclone V GX 5CGXFC5C6F27C7N FPGA, whose dedicated DSP blocks and embedded M10K memory resources are well-suited to the pipelined FFT and FIR workloads required. Hardware resource analysis shows that the complete reconstruction pipeline (comprising of a UART receive buffer, symmetric FIR filter, row-wise IFFT, matrix transposition, and column-wise IFFT) consumes fewer than 12 of the 150 available DSP blocks across target resolutions ranging from 64×64 to 256×256, demonstrating substantial headroom for future extension.

Communication to the host device is achieved through an Adafruit FT232H breakout board over GPIO, providing up to 1.2 MB/s throughput in Virtual COM Port (VCP) mode via MATLAB’s serialport() function. The reconstructed output is evaluated against a reference image produced in MATLAB software, using Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) to assess the effect of hardware filtering and acceleration on image quality.

---

Table of Contents

This table of contents serves as a visualisation of the topics covered in the interim report of my final year project. Please contact me if you wish to read the full paper.

• Abstract
• Acknowledgements
• Contributions
• Table of Contents
• List of Figures
• List of Tables
• Nomenclature
• Chapter 1: Introduction
  • 1.1 Project Motivation
    • 1.1.1 Latency in Reconstruction
    • 1.1.2 Limitations of General-Purpose Computing
  • 1.2 Project Objective
• Chapter 2: Literature Review
  • 2.1 Overview of the MRI Reconstruction Pipeline
  • 2.2 MRI Data Acquisition
    • 2.2.1 Sampling Trajectory
      • 2.2.1.1 Cartesian
      • 2.2.1.2 Radial
      • 2.2.1.3 Spiral
    • 2.2.2 K-Space Representation
    • 2.2.3 Frequency Sampling (Over/Under)
  • 2.3 Image Processing
    • 2.3.1 Processing in K-Space
    • 2.3.2 Quantitative Spatial Image Analysis
      • 2.3.2.1 PSNR
      • 2.3.2.2 Structural Similarity Index (SSIM)
  • 2.4 Image Reconstruction
    • 2.4.1 Fast Fourier Transform (FFT)
    • 2.4.2 Non-Uniform Fast Fourier Transform (NUFFT)
• Chapter 3: FPGA Architecture
  • 3.1 Overview
  • 3.2 Interconnects
    • 3.2.1 Row / Column Interconnect
    • 3.2.2 Local Interconnect
    • 3.2.3 Direct Link Interconnect
  • 3.3 Memory Architecture
  • 3.4 Lookup Tables (LUTs)
  • 3.5 Intellectual Property (IP) Blocks
  • 3.6 Digital Signal Processing (DSP) Block
    • 3.6.1 FFT and FIR Filtering in Hardware
• Chapter 4: Software Architecture
  • 4.1 Overview
  • 4.2 Verilog HDL
    • 4.2.1 Behavioural and Structural Descriptions
  • 4.3 Development Environments
  • 4.4 Data Interfacing with FPGA
• Chapter 5: System Design
  • 5.1 Development Workflow
    • 5.1.1 IP Core Access & Licensing
  • 5.2 Communication Interfaces
    • 5.2.1 Option A: UART
    • 5.2.2 Option B: UART (Reduced Dataset) and JTAG
    • 5.2.3 Option C: PCIe 2.0 (Hard IP) & HSMC
    • 5.2.4 Option D: Parallel Serial Communications through GPIO
    • 5.2.5 Discussion
  • 5.3 Resource Requirements and Throughput Analysis
    • 5.3.1 FFT IP Core Resource Characterisation
    • 5.3.2 Two-Dimensional Reconstruction Resource Budget
    • 5.3.3 Throughput and Frame Rate Analysis
  • 5.4 Pre-Processing: K-Space FIR Filter Design
  • 5.5 Pipelining Strategy
  • 5.6 MATLAB Reference and Evaluation
• Chapter 6: Conclusion
  • 6.1 Proposed Timeline for Implementation
• References