Deep unsupervised learning for accelerated MRI reconstruction

Simple item page
buir.advisorÇukur, Tolga
dc.contributor.authorKorkmaz, Yılmaz
dc.date.accessioned2022-08-26T11:31:03Z
dc.date.available2022-08-26T11:31:03Z
dc.date.copyright2022-07
dc.date.issued2022-07
dc.date.submitted2022-08-25
dc.descriptionCataloged from PDF version of article.en_US
dc.descriptionThesis (Master's): Bilkent University, Department of Electrical and Electronics Engineering, İhsan Doğramacı Bilkent University, 2022.en_US
dc.descriptionIncludes bibliographical references (leaves 45-71).en_US
dc.description.abstractSupervised reconstruction models are characteristically trained on matched pairs of undersampled and fully-sampled data to capture an MRI prior, along with supervision regarding the imaging operator to enforce data consistency. To reduce supervision requirements, the recent deep image prior framework instead conjoins untrained MRI priors with the imaging operator during inference. Yet, canonical convolutional architectures are suboptimal in capturing long-range relationships, and priors based on randomly initialized networks may yield suboptimal performance. To address these limitations, this thesis introduces a novel unsupervised MRI reconstruction method based on zero-Shot Learned Adversarial TransformERs (SLATER). SLATER embodies a deep adversarial network with cross-attention transformers to map noise and latent variables onto coil-combined MR images. During pre-training, this unconditional network learns a high-quality MRI prior in an unsupervised generative modeling task. During inference, a zero-shot reconstruction is then performed by incorporating the imaging operator and optimizing the prior to maximize consistency to undersampled data. Comprehensive experiments on brain MRI datasets clearly demonstrate the superior performance of SLATER against state-of-the-art unsupervised methods.en_US
dc.description.provenanceSubmitted by Betül Özen (ozen@bilkent.edu.tr) on 2022-08-26T11:31:03Z No. of bitstreams: 1 B161215.pdf: 48415047 bytes, checksum: eda0efb218258a4fdee20a56c633abf0 (MD5)en
dc.description.provenanceMade available in DSpace on 2022-08-26T11:31:03Z (GMT). No. of bitstreams: 1 B161215.pdf: 48415047 bytes, checksum: eda0efb218258a4fdee20a56c633abf0 (MD5) Previous issue date: 2022-07en
dc.description.statementofresponsibilityby Yılmaz Korkmazen_US
dc.format.extentxiii, 91 leaves : color illustrations, charts, graphics ; 30 cm.en_US
dc.identifier.itemidB161215
dc.identifier.urihttp://hdl.handle.net/11693/110473
dc.language.isoEnglishen_US
dc.rightsinfo:eu-repo/semantics/openAccessen_US
dc.subjectAdversarialen_US
dc.subjectTransformersen_US
dc.subjectMRIen_US
dc.subjectUnsuperviseden_US
dc.subjectReconstructionen_US
dc.subjectZero-shoten_US
dc.subjectGenerativeen_US
dc.titleDeep unsupervised learning for accelerated MRI reconstructionen_US
dc.title.alternativeDerin denetimsiz öğrenme ile hızlandırılmış MRG rekonstrüksiyonuen_US
dc.typeThesisen_US
thesis.degree.disciplineElectrical and Electronic Engineering
thesis.degree.grantorBilkent University
thesis.degree.levelMaster's
thesis.degree.nameMS (Master of Science)

Files

Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
B161215.pdf
Size:
46.17 MB
Format:
Adobe Portable Document Format
Description:
Full printable version
Download
License bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
license.txt
Size:
1.69 KB
Format:
Item-specific license agreed upon to submission
Description:
Download

Collections

Graduate School of Engineering and Science