Eshaghzadeh Torbati, Mahbaneh
(2024)
Harmonization of multi-scanner magnetic resonance imaging data.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
The integration of datasets from multiple sites or scanners in neuroimaging studies has become increasingly prevalent. However, the presence of substantial technical variability associated with scanners poses a challenge that can introduce unintended biases in downstream analyses. Moreover, this scanner-related variability, known as scanner effects, can manifest in longitudinal neuroimaging data due to potential scanner upgrades or replacements at sites. Harmonization methods have emerged as techniques to address scanner effects on multi-scanner neuroimaging data, encompassing both brain images and image-derived summary measures. Harmonization can be accomplished through various approaches, including the estimation and removal of scanner effects, as well as adapting the multi-scanner data to a scanner-middle-ground space or a target scanner domain. In these approaches, matched data can serve as additional labeled dataset to uncover scanner effects in the multi-scanner data. Harmonization methods that utilize matched data are referred to as supervised harmonization methods, leading many sites to collect additional matched data to facilitate harmonization. However, the current availability of neuroimaging data often lacks such matched data. Consequently, a thorough understanding of scanner effects and the development of both supervised and unsupervised harmonization methods are imperative.
This dissertation contributes to the field of harmonization of T1-weighted MRIs in several ways. Firstly, scanner effects and two harmonization methods for mitigating scanner effects in both images and image-derived measures are investigated. Secondly, MISPEL, a novel supervised image harmonization method is developed. MISPEL leverages matched data to learn a mapping to a scanner-middle-ground space. Third, a novel unsupervised image harmonization method, ESPA, is proposed. ESPA simulates scanner effects as augmentations on images and learns to harmonize images by adapting them to a scanner-middle-ground space. These contributions aim to enhance the understanding and effectiveness of harmonization techniques, addressing the challenges posed by scanner effects in neuroimaging studies.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
|
| Date: |
29 August 2024 |
| Date Type: |
Publication |
| Defense Date: |
11 June 2024 |
| Approval Date: |
29 August 2024 |
| Submission Date: |
23 June 2024 |
| Access Restriction: |
No restriction; Release the ETD for access worldwide immediately. |
| Number of Pages: |
158 |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
School of Computing and Information > Intelligent Systems Program |
| Degree: |
PhD - Doctor of Philosophy |
| Thesis Type: |
Doctoral Dissertation |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
Harmonization, Scanner effects, Deep learning |
| Date Deposited: |
29 Aug 2024 20:03 |
| Last Modified: |
29 Aug 2024 20:03 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/46574 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}