Cardiff University | Prifysgol Caerdydd ORCA
Online Research @ Cardiff 
WelshClear Cookie - decide language by browser settings

Efficient parametric imaging with GPU computing

Zhang, D., Zhu, X., Bifone, A., Gozzi, A., Capuani, S. and Palombo, M. ORCID: https://orcid.org/0000-0003-4892-7967 2017. Efficient parametric imaging with GPU computing. Biophysical Journal 112 (3) , 583A-584A. 10.1016/j.bpj.2016.11.3141

Full text not available from this repository.

Abstract

Parametric imaging plays a critical role in modern biophysical, biomedical research and clinical diagnosis. It can provide useful visual representations of a sample with respect to the parameters underlying the mathematical models associated with sample data. For this technique, nonlinear model fitting optimization is a commonly used approach to estimate the parameters on a pixel-by-pixel basis to create parametric maps. With the increased sophistication of modern imaging systems, the amount of data processed with parametric imaging techniques is exploding and the processing time is often limiting the advancement of these technologies for real-time and automotive applications. However, in many applications, the computation for each image pixel can be carried out independently of any others, and such sort of computations can profit tremendously from parallel processing. Nowadays, graphics processing unit (GPU) has become a standard tool in high-performance parallel computing. To realize real-time automated image reconstruction for parametric imaging techniques, such as multi-parametric microscopy and magnetic resonance imaging (MRI), we have developed a GPU-based nonlinear model fitting optimizer called GPU-LMFit. We demonstrate the applications of GPU-LMFit in super resolution localization microscopy, fluorescence lifetime imaging microscopy, diffusion-weighted MRI (DW-MRI) and myocardial longitudinal relaxation time (T1) MRI using modified Look-Locker inversion recovery (MOLLI) based techniques. The results show that the use of GPU-LMFit can readily result in more than tens of times of speedup of parametric analyses in these techniques, compared with the software using CPU-only processing. An important example will be presented that when GPU-LMFit was used with a medium level GPU like Quadro K2000 for a DW-MRI image data set to reconstruct non-Gaussian diffusion parametric images, the results show that the images can be constructed up to 240x faster than with CPU processing alone. In this application, GPU-LMFit helps to reduce the time for DW-MRI processing from hours to seconds. Our results show the performance of GPU-LMFit is excellent to significantly improve the efficiency of parametric analyses and can thus be a useful tool to enable automated parametric imaging for real-time visualization, analysis and diagnostics.

Item Type: Article
Date Type: Published Online
Status: Published
Schools: Psychology
Publisher: Biophysical Society
ISSN: 0006-3495
Last Modified: 10 Nov 2022 10:43
URI: https://orca.cardiff.ac.uk/id/eprint/147894

Actions (repository staff only)

Edit Item Edit Item