東華大學圖書館 |

A Machine Learning Inversion Framework for Brain Magnetic Resonance Elastography.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	A Machine Learning Inversion Framework for Brain Magnetic Resonance Elastography./
作者:	Scott, Jonathan Michael.
面頁冊數:	1 online resource (143 pages)
附註:	Source: Dissertations Abstracts International, Volume: 83-12, Section: B.
Contained By:	Dissertations Abstracts International83-12B.
標題:	Medical imaging. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28542940click for full text (PQDT)
ISBN:	9798802751411

A Machine Learning Inversion Framework for Brain Magnetic Resonance Elastography.
Scott, Jonathan Michael.

A Machine Learning Inversion Framework for Brain Magnetic Resonance Elastography. - 1 online resource (143 pages)

Source: Dissertations Abstracts International, Volume: 83-12, Section: B.

Thesis (Ph.D.)--College of Medicine - Mayo Clinic, 2023.

Includes bibliographical references

Magnetic resonance elastography (MRE) is a technique for the noninvasive measurement of tissue material properties. In MRE, shear waves are induced into a tissue of interest using an external vibration source and the wave propagation is imaged using a specialized MRI pulse sequence. An inversion algorithm is then applied to the acquired wave data to generate quantitative maps of the tissue material properties. Many of these inversion algorithms have intrinsic assumptions, such as local homogeneity and the presence of reliable data in all voxels of their processing footprint, that allow them to work very well for some applications but produce bias in others. MRE of the brain is a setting where these assumptions are often inaccurate across much of the imaged volume, and lead to significant bias in material property estimates. The hypothesis tested in this thesis is that a combination of wave physics simulations and machine learning algorithms can be used to generate an inversion framework that does not have the same limitations, and as a result provides improved results in settings where those assumptions are inaccurate.The first part of this thesis details the implementation of two models for simulating MRE-like wave data: coupled harmonic oscillators (CHO) and a finite difference model (FDM) of linear elasticity. Next, wave simulations in inhomogeneous materials generated with the CHO model are used to train an artificial neural network to return stiffness estimates. This inhomogeneous learned inversion (ILI) is then evaluated in a series of simulation, phantom, and in vivo experiments. In simulation, ILI provided more accurate and sharper resolution stiffness estimates than two inversions assuming material homogeneity while exhibiting little bias in noisy data. ILI also provided higher contrast to noise ratio in a gel phantom simulating brain with stiff inclusions, and higher effective resolution in vivo in a series of stiff meningiomas. In the final experimental chapter of this thesis, an updated ILI framework is developed using the FDM to generate wave data in inhomogeneous materials to serve as training data for the neural network. In this implementation, partial volumed voxels and voxels with no data are included in the training process to model the situation encountered by inversions at the edges of the brain. The ILI inversion is then compared against a homogeneous learned inversion and is shown to be less biased by edge-of-brain effects in brain MRE simulation experiments, though it has reduced sensitivity to true changes in stiffness in the cortex of the simulations. This version of ILI is then applied in vivo in a study of aging, where significant decreases in stiffness with increasing age are reported in cortical regions of interest. The work in this thesis demonstrates that the developed ILI framework increases the effective resolution of stiffness estimates and reduces bias caused by volume changes and partial volumed voxels in the cortex. This should motivate further development of this framework for use in brain MRE and other MRE settings where the assumptions of existing inversion algorithms may be limiting performance.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2023

Mode of access: World Wide Web

ISBN: 9798802751411Subjects--Topical Terms:

3172799
Medical imaging.
Subjects--Index Terms:

ElastographyIndex Terms--Genre/Form:

542853
Electronic books.

A Machine Learning Inversion Framework for Brain Magnetic Resonance Elastography.
LDR:04557nmm a2200373K 4500 001 2362494
005 20231102121805.5
006 m o d
007 cr mn ---uuuuu
008 241011s2023 xx obm 000 0 eng d
020 $a 9798802751411
035 $a (MiAaPQ)AAI28542940
035 $a AAI28542940
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Scott, Jonathan Michael. $3 3703224
245 1 2 $a A Machine Learning Inversion Framework for Brain Magnetic Resonance Elastography.
264 0 $c 2023
300 $a 1 online resource (143 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 83-12, Section: B.
500 $a Advisor: Murphy, Matthew C.;Ehman, Richard L.
502 $a Thesis (Ph.D.)--College of Medicine - Mayo Clinic, 2023.
504 $a Includes bibliographical references
520 $a Magnetic resonance elastography (MRE) is a technique for the noninvasive measurement of tissue material properties. In MRE, shear waves are induced into a tissue of interest using an external vibration source and the wave propagation is imaged using a specialized MRI pulse sequence. An inversion algorithm is then applied to the acquired wave data to generate quantitative maps of the tissue material properties. Many of these inversion algorithms have intrinsic assumptions, such as local homogeneity and the presence of reliable data in all voxels of their processing footprint, that allow them to work very well for some applications but produce bias in others. MRE of the brain is a setting where these assumptions are often inaccurate across much of the imaged volume, and lead to significant bias in material property estimates. The hypothesis tested in this thesis is that a combination of wave physics simulations and machine learning algorithms can be used to generate an inversion framework that does not have the same limitations, and as a result provides improved results in settings where those assumptions are inaccurate.The first part of this thesis details the implementation of two models for simulating MRE-like wave data: coupled harmonic oscillators (CHO) and a finite difference model (FDM) of linear elasticity. Next, wave simulations in inhomogeneous materials generated with the CHO model are used to train an artificial neural network to return stiffness estimates. This inhomogeneous learned inversion (ILI) is then evaluated in a series of simulation, phantom, and in vivo experiments. In simulation, ILI provided more accurate and sharper resolution stiffness estimates than two inversions assuming material homogeneity while exhibiting little bias in noisy data. ILI also provided higher contrast to noise ratio in a gel phantom simulating brain with stiff inclusions, and higher effective resolution in vivo in a series of stiff meningiomas. In the final experimental chapter of this thesis, an updated ILI framework is developed using the FDM to generate wave data in inhomogeneous materials to serve as training data for the neural network. In this implementation, partial volumed voxels and voxels with no data are included in the training process to model the situation encountered by inversions at the edges of the brain. The ILI inversion is then compared against a homogeneous learned inversion and is shown to be less biased by edge-of-brain effects in brain MRE simulation experiments, though it has reduced sensitivity to true changes in stiffness in the cortex of the simulations. This version of ILI is then applied in vivo in a study of aging, where significant decreases in stiffness with increasing age are reported in cortical regions of interest. The work in this thesis demonstrates that the developed ILI framework increases the effective resolution of stiffness estimates and reduces bias caused by volume changes and partial volumed voxels in the cortex. This should motivate further development of this framework for use in brain MRE and other MRE settings where the assumptions of existing inversion algorithms may be limiting performance.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2023
538 $a Mode of access: World Wide Web
650 4 $a Medical imaging. $3 3172799
650 4 $a Biomedical engineering. $3 535387
653 $a Elastography
653 $a Machine learning
653 $a Magnetic resonance elastography
653 $a Wave physics
655 7 $a Electronic books. $2 lcsh $3 542853
690 $a 0574
690 $a 0541
710 2 $a ProQuest Information and Learning Co. $3 783688
710 2 $a College of Medicine - Mayo Clinic. $b Biomedical Engineering. $3 3178816
773 0 $t Dissertations Abstracts International $g 83-12B.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28542940 $z click for full text (PQDT)