Multiscale unbiased diffeomorphic atlas construction on Multi-GPUs

Linh Ha, Jens Krüger, Sarang Joshi, Cláudio T. Silva

Research output: Chapter in Book/Report/Conference proceedingChapter

Abstract

This chapter presents a high-performance multiscale 3D image-processing framework to exploit the parallel processing power of multiple graphic processing units (multi-GPUs) for medical image analysis. The construction of population atlases plays a central role in medical image analysis, particularly in understanding the variability of brain anatomy. The method projects a large set of images to a common coordinate system, creating a statistical average model of the population, and doing regression analysis of anatomical structures. The brain atlas construction is a powerful technique to study the physiology, evolution, and development of the brain, as well as disease progression. Two desired properties of the atlas construction are that it should be diffeomorphic and nonbiased. GPUs algorithms and data structures can be applied to a wide range of 3D image-processing applications and efficiently exploit the computational power and massive bandwidth offered by modern GPUs. The framework helps scientists solve computationally intensive problems that previously required supercomputing power. This chapter illustrates that it is possible to implement unbiased greedy iterative atlas construction on multi-GPUs. Also the framework allows implementation of more sophisticated registration problems, such as LDDMM, metamorphosis, or image current. Although each technique has a different trade-offbetween quality of results and the computation involved, the framework is capable of quantifying those trade-offs to suggest a good solution for the practical problem suitable with inputs and the accessible computational power. © 2011

Original languageEnglish (US)
Title of host publicationGPU Computing Gems Emerald Edition
PublisherElsevier Inc.
Pages771-791
Number of pages21
ISBN (Print)9780123849885
DOIs
StatePublished - 2011

Fingerprint

Brain
Image analysis
Image processing
Physiology
Regression analysis
Data structures
Bandwidth
Graphics processing unit
Processing
Statistical Models

ASJC Scopus subject areas

  • Computer Science(all)

Cite this

Ha, L., Krüger, J., Joshi, S., & Silva, C. T. (2011). Multiscale unbiased diffeomorphic atlas construction on Multi-GPUs. In GPU Computing Gems Emerald Edition (pp. 771-791). Elsevier Inc.. https://doi.org/10.1016/B978-0-12-384988-5.00048-6

Multiscale unbiased diffeomorphic atlas construction on Multi-GPUs. / Ha, Linh; Krüger, Jens; Joshi, Sarang; Silva, Cláudio T.

GPU Computing Gems Emerald Edition. Elsevier Inc., 2011. p. 771-791.

Research output: Chapter in Book/Report/Conference proceedingChapter

Ha, L, Krüger, J, Joshi, S & Silva, CT 2011, Multiscale unbiased diffeomorphic atlas construction on Multi-GPUs. in GPU Computing Gems Emerald Edition. Elsevier Inc., pp. 771-791. https://doi.org/10.1016/B978-0-12-384988-5.00048-6
Ha L, Krüger J, Joshi S, Silva CT. Multiscale unbiased diffeomorphic atlas construction on Multi-GPUs. In GPU Computing Gems Emerald Edition. Elsevier Inc. 2011. p. 771-791 https://doi.org/10.1016/B978-0-12-384988-5.00048-6
Ha, Linh ; Krüger, Jens ; Joshi, Sarang ; Silva, Cláudio T. / Multiscale unbiased diffeomorphic atlas construction on Multi-GPUs. GPU Computing Gems Emerald Edition. Elsevier Inc., 2011. pp. 771-791
@inbook{0b4e2160cb5641388ecb3469b2141511,
title = "Multiscale unbiased diffeomorphic atlas construction on Multi-GPUs",
abstract = "This chapter presents a high-performance multiscale 3D image-processing framework to exploit the parallel processing power of multiple graphic processing units (multi-GPUs) for medical image analysis. The construction of population atlases plays a central role in medical image analysis, particularly in understanding the variability of brain anatomy. The method projects a large set of images to a common coordinate system, creating a statistical average model of the population, and doing regression analysis of anatomical structures. The brain atlas construction is a powerful technique to study the physiology, evolution, and development of the brain, as well as disease progression. Two desired properties of the atlas construction are that it should be diffeomorphic and nonbiased. GPUs algorithms and data structures can be applied to a wide range of 3D image-processing applications and efficiently exploit the computational power and massive bandwidth offered by modern GPUs. The framework helps scientists solve computationally intensive problems that previously required supercomputing power. This chapter illustrates that it is possible to implement unbiased greedy iterative atlas construction on multi-GPUs. Also the framework allows implementation of more sophisticated registration problems, such as LDDMM, metamorphosis, or image current. Although each technique has a different trade-offbetween quality of results and the computation involved, the framework is capable of quantifying those trade-offs to suggest a good solution for the practical problem suitable with inputs and the accessible computational power. {\circledC} 2011",
author = "Linh Ha and Jens Kr{\"u}ger and Sarang Joshi and Silva, {Cl{\'a}udio T.}",
year = "2011",
doi = "10.1016/B978-0-12-384988-5.00048-6",
language = "English (US)",
isbn = "9780123849885",
pages = "771--791",
booktitle = "GPU Computing Gems Emerald Edition",
publisher = "Elsevier Inc.",

}

TY - CHAP

T1 - Multiscale unbiased diffeomorphic atlas construction on Multi-GPUs

AU - Ha, Linh

AU - Krüger, Jens

AU - Joshi, Sarang

AU - Silva, Cláudio T.

PY - 2011

Y1 - 2011

N2 - This chapter presents a high-performance multiscale 3D image-processing framework to exploit the parallel processing power of multiple graphic processing units (multi-GPUs) for medical image analysis. The construction of population atlases plays a central role in medical image analysis, particularly in understanding the variability of brain anatomy. The method projects a large set of images to a common coordinate system, creating a statistical average model of the population, and doing regression analysis of anatomical structures. The brain atlas construction is a powerful technique to study the physiology, evolution, and development of the brain, as well as disease progression. Two desired properties of the atlas construction are that it should be diffeomorphic and nonbiased. GPUs algorithms and data structures can be applied to a wide range of 3D image-processing applications and efficiently exploit the computational power and massive bandwidth offered by modern GPUs. The framework helps scientists solve computationally intensive problems that previously required supercomputing power. This chapter illustrates that it is possible to implement unbiased greedy iterative atlas construction on multi-GPUs. Also the framework allows implementation of more sophisticated registration problems, such as LDDMM, metamorphosis, or image current. Although each technique has a different trade-offbetween quality of results and the computation involved, the framework is capable of quantifying those trade-offs to suggest a good solution for the practical problem suitable with inputs and the accessible computational power. © 2011

AB - This chapter presents a high-performance multiscale 3D image-processing framework to exploit the parallel processing power of multiple graphic processing units (multi-GPUs) for medical image analysis. The construction of population atlases plays a central role in medical image analysis, particularly in understanding the variability of brain anatomy. The method projects a large set of images to a common coordinate system, creating a statistical average model of the population, and doing regression analysis of anatomical structures. The brain atlas construction is a powerful technique to study the physiology, evolution, and development of the brain, as well as disease progression. Two desired properties of the atlas construction are that it should be diffeomorphic and nonbiased. GPUs algorithms and data structures can be applied to a wide range of 3D image-processing applications and efficiently exploit the computational power and massive bandwidth offered by modern GPUs. The framework helps scientists solve computationally intensive problems that previously required supercomputing power. This chapter illustrates that it is possible to implement unbiased greedy iterative atlas construction on multi-GPUs. Also the framework allows implementation of more sophisticated registration problems, such as LDDMM, metamorphosis, or image current. Although each technique has a different trade-offbetween quality of results and the computation involved, the framework is capable of quantifying those trade-offs to suggest a good solution for the practical problem suitable with inputs and the accessible computational power. © 2011

UR - http://www.scopus.com/inward/record.url?scp=84884437423&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84884437423&partnerID=8YFLogxK

U2 - 10.1016/B978-0-12-384988-5.00048-6

DO - 10.1016/B978-0-12-384988-5.00048-6

M3 - Chapter

SN - 9780123849885

SP - 771

EP - 791

BT - GPU Computing Gems Emerald Edition

PB - Elsevier Inc.

ER -