- Poster presentation
- Open Access
Accelerated cloud and GPU-based simulations for quantification of relaxation times: an example with MOLLI
© Kantasis et al. 2016
- Published: 27 January 2016
- Execution Time
- Master Node
- Total Execution Time
- Slave Node
- Spin Population
Quantification of native T1 in the myocardium has the potential to become an important biomarker for assessing specific cardiomyopathies. However, the accuracy and precision of the clinically available CMR mapping techniques are affected by several parameters .
A recently developed method that allows for accuracy improvement of MOLLI quantification was presented [2, 3]. This method is based on comprehensive, GPU-based MR simulations of the identical pulse sequence on a large population of spins resulting in a database of all possible outcomes. MR simulations are computationally intensive with long execution times not well suited for clinical applications. The aim of this study was to improve performance to more clinically acceptable levels. We hypothesized that this could be accomplished by a cloud and GPU-based implementation approach.
An Amazon Web Services  cloud-based cluster consisting of g2.2 × large computer instances was utilized. A variable number of instances were used as slave nodes. Each node performed the simulation of the entire pulse sequence on a subset of the spin population and one of them was also assigned the role of job manager (master node). The resulting database entries were then collected from the slave nodes and joined together on the master node.
The simulated pulse sequence was a MOLLI with a 5(3p)3 acquisition scheme consisting of 156877 discrete time-steps. The spin population consisted of 2756061 spins with variable T1 and T2 combinations, covering physiological myocardial and blood relaxation times.
The execution times were measured and the total speedup was calculated for 1 to 16 nodes. The total speedup was calculated according to the total execution time and the GPU speedup according to solely the execution time of the GPU part of the simulation. The overhead, defined as the time of data transfers, joining the database and preparing to run the simulation, was also measured.
In this study, a cloud-based approach of simulation-based corrections on MOLLI T1 mapping data [2, 3] was used. The measured speedups reflect the benefits of distributing the computational problem across multiple nodes within the cloud. This work suggests that, in the future, using the cloud may allow for simulation-based correction of patient data to become useful in the clinic.
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.