Bilkent University Institutional Repository (BUIR)

Bilkent University Institutional Repository (BUIR), a service of Bilkent University Libraries, collects, preserves, and distributes the intellectual output of Bilkent University. Faculty, staff, and students are invited to deposit their research and scholarship. Departments, administrative units, programs, and centers are invited to use the Institutional Repository to distribute their working papers, technical reports, conference proceedings, and other research material. For assistance in depositing documents. For more information, please contact us.

 
openaccess

Scholarly Publications

openaccess

University Library

openaccess

Bilkent Theses

Recent Submissions

ItemOpen Access
Unsupervised medical image translation with adversarial diffusion models
(Institute of Electrical and Electronics Engineers Inc., 2023-12-01) Özbey, Muzaffer; Dalmaz, Onat; Dar, Salman U. H.; Bedel, Hasan A.; Özturk, Şaban; Güngör, Alper; Çukur, Tolga
Imputation of missing images via source-to-target modality translation can improve diversity in medical imaging protocols. A pervasive approach for synthesizing target images involves one-shot mapping through generative adversarial networks (GAN). Yet, GAN models that implicitly characterize the image distribution can suffer from limited sample fidelity. Here, we propose a novel method based on adversarial diffusion modeling, SynDiff, for improved performance in medical image translation. To capture a direct correlate of the image distribution, SynDiff leverages a conditional diffusion process that progressively maps noise and source images onto the target image. For fast and accurate image sampling during inference, large diffusion steps are taken with adversarial projections in the reverse diffusion direction. To enable training on unpaired datasets, a cycle-consistent architecture is devised with coupled diffusive and non-diffusive modules that bilaterally translate between two modalities. Extensive assessments are reported on the utility of SynDiff against competing GAN and diffusion models in multi-contrast MRI and MRI-CT translation. Our demonstrations indicate that SynDiff offers quantitatively and qualitatively superior performance against competing baselines.
ItemOpen Access
Scaling stratified stochastic gradient descent for distributed matrix completion
(IEEE Computer Society, 2023-10-01) Abubaker, Nabil; Karsavuran, M. O.; Aykanat, Cevdet
Stratified SGD (SSGD) is the primary approach for achieving serializable parallel SGD for matrix completion. State-of-the-art parallelizations of SSGD fail to scale due to large communication overhead. During an SGD epoch, these methods send data proportional to one of the dimensions of the rating matrix. We propose a framework for scalable SSGD through significantly reducing the communication overhead via exchanging point-to-point messages utilizing the sparsity of the rating matrix. We provide formulas to represent the essential communication for correctly performing parallel SSGD and we propose a dynamic programming algorithm for efficiently computing them to establish the point-to-point message schedules. This scheme, however, significantly increases the number of messages sent by a processor per epoch from O(K) to (K2) for a K-processor system which might limit the scalability. To remedy this, we propose a Hold-and-Combine strategy to limit the upper-bound on the number of messages sent per processor to O(KlgK). We also propose a hypergraph partitioning model that correctly encapsulates reducing the communication volume. Experimental results show that the framework successfully achieves a scalable distributed SSGD through significantly reducing the communication overhead. Our code is publicly available at: github.com/nfabubaker/CESSGD
ItemOpen Access
Segmentatıon of spınal subarachnoıd lumen wıth 3d attentıon u net
(World Scientific Publishing, 2023-05-01) Keleş, A.; Algın, Oktay; Özışık Akdemir, P.; Şen, B.; Çelebi, F. V.
Phase Contrast Magnetic Resonance Image (PC-MRI) is an emerging noninvasive technique that contains pulsatile information by measuring the parameters of cerebrospinal fluid (CSF) flow. As CSF flow quantities are measured from the selected region on the images, the accuracy in the identification of the interested region is the most essential, and the examination requires a lot of time and experience to analyze and for accurate CSF flow assessment. In this study, a three-dimensional (3D)-Unet architecture, including pulsatile flow data as the third dimension, is proposed to address the issue. The dataset contains 2176 phase and rephase images from 57 slabs of 39 3-tesla PC-MRI subjects collected from the lower thoracic levels of control and Idiopathic Scoliosis (IS) patients. The procedure starts with labeling the CSF containing spaces in the spinal canal. In the preprocessing step, unequal cardiac cycle images (i.e., frame) and the numbers of MRIs in cases are adjusted by interpolation to align the temporal dimension of the dataset to an equal size. The five-fold cross-validation procedure is used to evaluate the 3D Attention-U-Net model after training and achieved an average weighted performance of 97% precision, 95% recall, 98% F1 score, and 95% area under curve. The success of the model is also measured using the CSF flow waveform quantities as well. The mean flow rates through the labeled and predicted CSF lumens have a significant correlation coefficient of 0.96, and the peak CSF flow rates have a coefficient of 0.65. To our knowledge, this is the first fully automatic 3D deep learning architecture implementation to segment spinal CSF-containing spaces that utilizes both spatial and pulsatile information in PC-MRI data. We expect that our work will attract future research on the use of PC-MRI temporal information for training deep models.
ItemEmbargo
Complex patterning of matter with liesegang patterns propagating through different concentration media─gel lenses for liesegang waves
(American Chemical Society (ACS) Publications, 2023-11-18) Akbulut, Elif Sıla; Holló, Gábor; Lagzi, Istvan; Baytekin, Bilge
The patterns formed in natural biochemical and geochemical media are never spatially or geometrically homogeneous. On the other hand, the artificial systems trying to mimic nature are usually homogeneous and far from depicting the complexity of the natural ones. Liesegang patterns (LPs) are artificial reaction-diffusion precipitate patterns that can be formed in hydrogels. Although these patterns can be made to “sense” the environment, they are mostly formed in homogeneous media. Here, we present that a simple setting of different gel concentration boundaries can cause refractions of the pattern waves and changes in the band spacings. The extent of refraction is dependent on the macroscopic shape of the boundary. As imaged by scanning electron microscopy, the LP bands “crossing the boundaries” are formed by the product of a new morphology. This study can be a step forward in straightforwardly achieving complexity in artificial systems and developing new crystal forms of solids.
ItemOpen Access
Uncertainty principles in holomorphic function spaces on the unit ball
(Cambridge University Press, 2023-07-10) Kaptanoğlu, Hakkı Turgay
On all Bergman–Besov Hilbert spaces on the unit disk, we find self-adjoint weighted shift operators that are differential operators of half-order whose commutators are the identity, thereby obtaining uncertainty relations in these spaces. We also obtain joint average uncertainty relations for pairs of commuting tuples of operators on the same spaces defined on the unit ball. We further identify functions that yield equality in some uncertainty inequalities.