자료유형  

 학위논문

Control Number  

0017163722

International Standard Book Number  

9798342107495

Dewey Decimal Classification Number  

616.8905

Main Entry-Personal Name  

Ouyang, Jiahong.

Publication, Distribution, etc. (Imprint  

[S.l.] : Stanford University., 2024

Publication, Distribution, etc. (Imprint  

Ann Arbor : ProQuest Dissertations & Theses, 2024

Physical Description  

195 p.

General Note  

Source: Dissertations Abstracts International, Volume: 86-04, Section: B.

General Note  

Advisor: Pohl, Kilian;Zaharchuk, Greg.

Dissertation Note  

Thesis (Ph.D.)--Stanford University, 2024.

Summary, Etc.  

요약Multi-modal and longitudinal neuroimages (a.k.a. multi-dimensional neuroimages) are critical for the understanding, diagnosis, and monitoring of neurological disorders. The complex disease patterns captured by these images are in many cases difficult to identify by visual inspection from human experts or existing technology. Deep learning techniques have recently shown immense potential in neuroimage analysis. However, they often result in uninterpretable findings, which is of particular concern where understanding a model's behavior fosters trust and assurance among clinicians. Besides, to make their findings generalizable, they usually require large labeled neuroimaging datasets that are unavailable or acquired at a high cost. Thus, in this dissertation, we aim to address these two challenges of deep learning approaches: interpretability and accuracy under limited data. Specifically, we propose to enhance interpretability by visualization and estimation of patterns characteristic for a disease. To accurately identify disease-specific patterns, we propose to integrate prior knowledge in the model design and to develop novel deep learning strategies centered around self- or weakly supervision.Adapting these key ideas, we first develop deep learning methods for multi-modal neuroimages with the task of synthesizing 18F-fluorodeoxyglucose (FDG) Positron Emission Tomography (PET) from multi-contrast Magnetic Resonance Imaging (MRI). We introduce brain symmetry into the model design to achieve accurate characterization of abnormality. Then, we develop a self-supervised method to enable accurate synthesis even when an input modality is missing. We are able to synthesize diagnostic-quality FDG PET images from MRIs for the brain neoplasm cohort, potentially leading to safer and more equitable diagnostic neuroimaging. Secondly, we design a series of interpretable deep learning methods ranging from supervised to self- or weakly supervised to analyze brain aging and Alzheimer's Disease (AD) from longitudinal MRIs. These models explicitly account for the irreversibility of these processes enabling us to accurately estimate brain age and disease progression, including AD diagnoses and identifying subjects that will convert to AD. Lastly, we introduce a work that further extends to the interpretable analysis of multi-dimensional neuroimages, that jointly learns from longitudinal MRI and amyloid PET. By regularizing the temporal ordering of showing disease abnormality across modalities, it further results in the accurate cross-modal prediction task of estimating amyloid status from MRI. These efforts in AD analysis enable early-stage diagnosis of AD, which has the potential of facilitating timely intervention and enhancing AD clinical trials.

Subject Added Entry-Topical Term  

Neuroimaging.

Subject Added Entry-Topical Term  

Alzheimer's disease.

Subject Added Entry-Topical Term  

Deep learning.

Subject Added Entry-Topical Term  

Neurological disorders.

Subject Added Entry-Topical Term  

Aging.

Subject Added Entry-Topical Term  

Medical research.

Subject Added Entry-Topical Term  

Brain.

Subject Added Entry-Topical Term  

Medical imaging.

Subject Added Entry-Topical Term  

Visualization.

Subject Added Entry-Topical Term  

Medicine.

Subject Added Entry-Topical Term  

Neurosciences.

Added Entry-Corporate Name  

Stanford University.

Host Item Entry  

Dissertations Abstracts International. 86-04B.

Electronic Location and Access  

로그인을 한후 보실 수 있는 자료입니다.

Control Number  

joongbu:657569