語系:
繁體中文
English
說明(常見問題)
回圖書館首頁
手機版館藏查詢
登入
回首頁
切換:
標籤
|
MARC模式
|
ISBD
FindBook
Google Book
Amazon
博客來
Optimization-Based Deep Learning Methods for Magnetic Resonance Imaging Reconstruction and Synthesis.
紀錄類型:
書目-電子資源 : Monograph/item
正題名/作者:
Optimization-Based Deep Learning Methods for Magnetic Resonance Imaging Reconstruction and Synthesis./
作者:
Bian, Wanyu.
面頁冊數:
1 online resource (146 pages)
附註:
Source: Dissertations Abstracts International, Volume: 84-03, Section: B.
Contained By:
Dissertations Abstracts International84-03B.
標題:
Applied mathematics. -
電子資源:
http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=29211216click for full text (PQDT)
ISBN:
9798351426334
Optimization-Based Deep Learning Methods for Magnetic Resonance Imaging Reconstruction and Synthesis.
Bian, Wanyu.
Optimization-Based Deep Learning Methods for Magnetic Resonance Imaging Reconstruction and Synthesis.
- 1 online resource (146 pages)
Source: Dissertations Abstracts International, Volume: 84-03, Section: B.
Thesis (Ph.D.)--University of Florida, 2022.
Includes bibliographical references
This dissertation is devoted to provide advanced nonconvex nonsmooth variational model of (Magnetic Resonance Image) MRI reconstruction, efficient learnable image reconstruction algorithms and parameter training algorithms that improve the accuracy and robustness of the optimization-based deep learning methods for compressed sensing MRI reconstruction and synthesis.The first part and second part consider deep learning based optimization algorithms for parallel MRI (pMRI) reconstruction with significant undersampled data without coil sensitivities. The third and fourth part solve single coil MRI reconstruction problem with diverse dataset where the trained network has convergence guarantee in theory and the network is trained by bilevel optimization algorithm.The first part introduces a novel optimization based deep neural network whose architecture is inspired by proximal gradient descent for solving a variational model of the pMRI reconstruction problem without knowledge of coil sensitivity maps.Our variational model consists of data fidelity and regularization, the regularization function designated to be applied on channel-wise multi-coil images, which consists of the nonlinear combination operator and sparse feature encoder. The second part is a substantial extension of the preliminary work in the first part by solving the calibration-free fast pMRI reconstruction problem in a discrete-time optimal control framework. The network architecture is determined by the discrete-time dynamic system, which is induced by a designated variational model. The variational model consists of data fitting term and the regularization which contains two parts: one is for enhancing the sparsity of the reconstructed image in images domain; the other is in the k-space domain to further remove high-frequency artifacts.We demonstrate that the Lagrangian method for solving the control problem is equivalent to back-propagation. By learning multi-coil image combination operators and performing regularizations in both the image domain and k-space domain, the proposed method achieves a highly efficient image reconstruction network for pMRI.The third part aims at developing a generalizable Magnetic Resonance Imaging (MRI) reconstruction method in the meta-learning framework. Specifically, we developed a deep reconstruction network induced by a learnable optimization algorithm (LOA) to solve the nonconvex nonsmooth variational model of MRI image reconstruction. In this model, the nonconvex nonsmooth regularization term is parameterized as a structured deep network where the network parameters can be learned from data. We partition these network parameters into two parts: a task-invariant part for the common feature encoder component of the regularization, and a task-specific part to account for the variations in the heterogeneous training and testing data. We train the regularization parameters in a bilevel optimization framework which significantly improves the robustness of the training process and the generalization ability of the network. We conducted a series of numerical experiments using heterogeneous MRI data sets with various undersampling patterns, ratios, and acquisition settings. The experimental results show that our network yields greatly improved reconstruction quality over existing methods and can generalize well to new reconstruction problems whose undersampling patterns/trajectories are not present during training.The last part aims to synthesize target modality of MRI (e.g., proton density image) by using partially scanned k-space data from source modalities (e.g., T1-weighted and T2-weighted image) instead of fully scanned data that is used in the state-of-the-art multimodal synthesis. We propose to learn three modality-specific feature extraction operators, one for each of these three modalities. Then, we design regularizers of these images by combining these learned operators and a robust sparse feature selection operator ($l_{2,1}$ norm). To synthesize the target modality image using source modalities, we employ another feature-fusion operator which learns the mapping from the features that are generated from source modalities to the target modality. We present a novel LOA for solving the joint multimodal MRI reconstruction and synthesis model with theoretical analysis guarantee. The parameters and hyper-parameters of the network induced by our LOA are learned using a bilevel optimization algorithm robust parameter training. Extensive experiments on a variety of pMRI and MRI imaging data sets with different modalities validate the magnificent performance of the proposed deep models. The experimental results demonstrate that our methods yield improved reconstruction quality and provide practical significance for clinical application and research studies.
Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2023
Mode of access: World Wide Web
ISBN: 9798351426334Subjects--Topical Terms:
2122814
Applied mathematics.
Subjects--Index Terms:
Image reconstructionIndex Terms--Genre/Form:
542853
Electronic books.
Optimization-Based Deep Learning Methods for Magnetic Resonance Imaging Reconstruction and Synthesis.
LDR
:06222nmm a2200373K 4500
001
2357464
005
20230710070041.5
006
m o d
007
cr mn ---uuuuu
008
241011s2022 xx obm 000 0 eng d
020
$a
9798351426334
035
$a
(MiAaPQ)AAI29211216
035
$a
AAI29211216
040
$a
MiAaPQ
$b
eng
$c
MiAaPQ
$d
NTU
100
1
$a
Bian, Wanyu.
$3
3697994
245
1 0
$a
Optimization-Based Deep Learning Methods for Magnetic Resonance Imaging Reconstruction and Synthesis.
264
0
$c
2022
300
$a
1 online resource (146 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: 84-03, Section: B.
500
$a
Advisor: Chen, Yunmei.
502
$a
Thesis (Ph.D.)--University of Florida, 2022.
504
$a
Includes bibliographical references
520
$a
This dissertation is devoted to provide advanced nonconvex nonsmooth variational model of (Magnetic Resonance Image) MRI reconstruction, efficient learnable image reconstruction algorithms and parameter training algorithms that improve the accuracy and robustness of the optimization-based deep learning methods for compressed sensing MRI reconstruction and synthesis.The first part and second part consider deep learning based optimization algorithms for parallel MRI (pMRI) reconstruction with significant undersampled data without coil sensitivities. The third and fourth part solve single coil MRI reconstruction problem with diverse dataset where the trained network has convergence guarantee in theory and the network is trained by bilevel optimization algorithm.The first part introduces a novel optimization based deep neural network whose architecture is inspired by proximal gradient descent for solving a variational model of the pMRI reconstruction problem without knowledge of coil sensitivity maps.Our variational model consists of data fidelity and regularization, the regularization function designated to be applied on channel-wise multi-coil images, which consists of the nonlinear combination operator and sparse feature encoder. The second part is a substantial extension of the preliminary work in the first part by solving the calibration-free fast pMRI reconstruction problem in a discrete-time optimal control framework. The network architecture is determined by the discrete-time dynamic system, which is induced by a designated variational model. The variational model consists of data fitting term and the regularization which contains two parts: one is for enhancing the sparsity of the reconstructed image in images domain; the other is in the k-space domain to further remove high-frequency artifacts.We demonstrate that the Lagrangian method for solving the control problem is equivalent to back-propagation. By learning multi-coil image combination operators and performing regularizations in both the image domain and k-space domain, the proposed method achieves a highly efficient image reconstruction network for pMRI.The third part aims at developing a generalizable Magnetic Resonance Imaging (MRI) reconstruction method in the meta-learning framework. Specifically, we developed a deep reconstruction network induced by a learnable optimization algorithm (LOA) to solve the nonconvex nonsmooth variational model of MRI image reconstruction. In this model, the nonconvex nonsmooth regularization term is parameterized as a structured deep network where the network parameters can be learned from data. We partition these network parameters into two parts: a task-invariant part for the common feature encoder component of the regularization, and a task-specific part to account for the variations in the heterogeneous training and testing data. We train the regularization parameters in a bilevel optimization framework which significantly improves the robustness of the training process and the generalization ability of the network. We conducted a series of numerical experiments using heterogeneous MRI data sets with various undersampling patterns, ratios, and acquisition settings. The experimental results show that our network yields greatly improved reconstruction quality over existing methods and can generalize well to new reconstruction problems whose undersampling patterns/trajectories are not present during training.The last part aims to synthesize target modality of MRI (e.g., proton density image) by using partially scanned k-space data from source modalities (e.g., T1-weighted and T2-weighted image) instead of fully scanned data that is used in the state-of-the-art multimodal synthesis. We propose to learn three modality-specific feature extraction operators, one for each of these three modalities. Then, we design regularizers of these images by combining these learned operators and a robust sparse feature selection operator ($l_{2,1}$ norm). To synthesize the target modality image using source modalities, we employ another feature-fusion operator which learns the mapping from the features that are generated from source modalities to the target modality. We present a novel LOA for solving the joint multimodal MRI reconstruction and synthesis model with theoretical analysis guarantee. The parameters and hyper-parameters of the network induced by our LOA are learned using a bilevel optimization algorithm robust parameter training. Extensive experiments on a variety of pMRI and MRI imaging data sets with different modalities validate the magnificent performance of the proposed deep models. The experimental results demonstrate that our methods yield improved reconstruction quality and provide practical significance for clinical application and research studies.
533
$a
Electronic reproduction.
$b
Ann Arbor, Mich. :
$c
ProQuest,
$d
2023
538
$a
Mode of access: World Wide Web
650
4
$a
Applied mathematics.
$3
2122814
650
4
$a
Medical imaging.
$3
3172799
653
$a
Image reconstruction
653
$a
Magnetic resonance imaging
653
$a
Nonconvex nonsmooth regularization term
653
$a
Scanned k-space data
655
7
$a
Electronic books.
$2
lcsh
$3
542853
690
$a
0364
690
$a
0574
710
2
$a
ProQuest Information and Learning Co.
$3
783688
710
2
$a
University of Florida.
$b
Mathematics.
$3
3343201
773
0
$t
Dissertations Abstracts International
$g
84-03B.
856
4 0
$u
http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=29211216
$z
click for full text (PQDT)
筆 0 讀者評論
館藏地:
全部
電子資源
出版年:
卷號:
館藏
1 筆 • 頁數 1 •
1
條碼號
典藏地名稱
館藏流通類別
資料類型
索書號
使用類型
借閱狀態
預約狀態
備註欄
附件
W9479820
電子資源
11.線上閱覽_V
電子書
EB
一般使用(Normal)
在架
0
1 筆 • 頁數 1 •
1
多媒體
評論
新增評論
分享你的心得
Export
取書館
處理中
...
變更密碼
登入