Related Experiment Video
Updated: Mar 27, 2026
02:09
Multi-modal Pulmonary Imaging: Using Complementary Information from CT and Hyperpolarized 129Xe MRI to Evaluate Lung Structure-Function
Published on: April 12, 2024
1.2K
Efficient methods for implementation of multi-level nonrigid mass-preserving image registration on GPUs and
Nathan D Ellingwood1, Youbing Yin2, Matthew Smith3
1IIHR-Hydroscience & Engineering, The University of Iowa, Iowa City, IA 52242, United States.
Computer Methods and Programs in Biomedicine
|January 19, 2016
Summary
This study introduces a novel GPU-accelerated method for lung image registration, significantly reducing computation time. The Diffeomorphic Multi-level B-Spline Transform Composite (DMTC) method achieves faster and more accurate medical image analysis for population studies and clinical use.
Area of Science:
- Medical Imaging
- Computational Anatomy
- High-Performance Computing
Background:
- Accurate medical image registration is crucial for population-based studies and clinical applications.
- Existing methods often face computational and memory limitations, hindering efficiency.
Purpose of the Study:
- To present a novel, computation- and memory-efficient multi-level registration method for computed tomography (CT) lung images using graphics processing units (GPUs).
- To enable faster and more accurate image registration for both research and clinical settings.
Main Methods:
- Developed a Diffeomorphic Multi-level B-Spline Transform Composite (DMTC) method for nonrigid, mass-preserving registration of CT lung images on GPUs.
- Employed a hierarchy of B-Spline grids and a sum of squared tissue volume difference (SSTVD) criterion for mass preservation.
- Optimized for GPU implementation to address memory constraints and reduce computation through pre-computed values and minimal CPU-GPU communication.
Main Results:
- The GPU-based DMTC method demonstrated good agreement with CPU-based results, with an average normalized root mean square error (nRMS) of 0.044±0.015.
- Achieved significant performance speedups, with the GPU version being up to 112 times faster than single-threaded CPU and 11 times faster than multi-threaded CPU at the highest resolution.
- Reduced total registration time to 2.9 minutes on GPU, compared to 12.8 minutes (12-threaded CPU) and 112.5 minutes (single-threaded CPU).
Conclusions:
- The proposed GPU-based DMTC method significantly outperforms its multi-threaded CPU counterpart in terms of runtime.
- The method offers substantial improvements in registration speed, making it highly suitable for large-scale medical imaging research and clinical workflows.
- The GPU implementation framework is adaptable for other cost functions requiring derivative calculations.
