Related Experiment Video
Updated: Jul 20, 2025
07:13
Author Spotlight: An Efficient and Robust Software for Automated Fusion of Multiple Preclinical Imaging Modalities
Published on: October 27, 2023
1.2K
FPGA Implementation of Image Registration Using Accelerated CNN
Seda Guzel Aydin1, Hasan Şakir Bilge2
1Department of Electrical and Electronics Engineering, Bingol University, Bingol 12000, Turkey.
Sensors (Basel, Switzerland)
|July 29, 2023
Summary
This study introduces an FPGA-based CNN for faster ultrasound image registration, achieving 139x speedup over software methods with minimal accuracy loss for real-time medical imaging applications.
Area of Science:
- Medical Imaging
- Computer Vision
- Hardware Acceleration
Background:
- Accurate ultrasound image registration (IR) is vital for image-guided surgery.
- Convolutional Neural Network (CNN) IR offers speed advantages over traditional methods.
- General-purpose processors limit real-time CNN performance; Field Programmable Gate Arrays (FPGAs) offer acceleration.
Purpose of the Study:
- To develop an FPGA-based CNN for accelerated ultrasound image registration.
- To regress three rigid registration parameters efficiently.
Main Methods:
- Proposed an FPGA-based ultrasound IR CNN (FUIR-CNN).
- Utilized fixed-point data and parallel operations (unrolling, pipelining) for speed.
- Implemented and tested on xc7z020 and xcku5p FPGAs with three US datasets.
Main Results:
- Achieved 139x faster inference compared to software-based CNNs.
- Maintained negligible regression performance drop (<200 MHz clock frequency).
Conclusions:
- The end-to-end FPGA-accelerated CNN provides high-speed registration with minimal accuracy loss.
- Demonstrated potential for real-time medical imaging with reduced power consumption compared to CPUs.
