Scholarly Publications - UMRAM
Permanent URI for this collectionhttps://hdl.handle.net/11693/115674
Browse
Recent Submissions
Item Open Access Separating nanoparticle induced delays from relaxation time constant in TAURUS(Infinite Science Publishing, 2025) Gülsün, Sevil Dilge; Alpman, Aslı; Sarıtaş, Emine ÜlküMagnetic nanoparticles (MNPs) exhibit relaxation behavior, introducing a delay in their magnetization alignment.The TAURUS method enables simultaneous estimation of the signal delay and the effective relaxation time constant.In this study, we propose a method to separately estimate the MNP induced delay and the system induced delay.We then demonstrate that the MNP induced delay and the relaxation time constant estimated via TAURUS showdifferent trends, implying that they may capture different aspects of the MNP response.Item Open Access Learning-based MRI reconstruction via a state-space U-Net model(IEEE, 2025) Yavuz, Muhammet Talat; Özturk, Şaban; Kabaş, Bilal; Çukur, TolgaMagnetic Resonance Imaging (MRI) is a widely used medical imaging technique that provides high-contrast visualization of soft tissues without ionizing radiation. However, long scan times impose significant limitations on clinical workflows, reducing patient comfort and increasing healthcare costs. Accelerated MRI techniques aim to shorten scan times by acquiring fewer k-space samples; however, this leads to undersampling artifacts and a decline in image quality. MRI reconstruction techniques are widely used to mitigate these artifacts by generating high-quality images from undersampled k-space data. In this study, we introduce MambaCC, an innovative State-Space Model (SSM)-based method that effectively captures long-range dependencies with low computational requirements, enabling efficient MRI reconstruction. MambaCC operates on a U-shaped deep learning network, where its outputs are enhanced with channel-mixing blocks to account for the effects of artifacts at the channel level. The performance of the proposed method has been evaluated on the fastMRI dataset and compared against state-of-the-art methods, including MoDL, U-Net, TransUNet, and UMamba. The results demonstrate that our approach achieves superior performance in terms of PSNR and SSIM metrics. The findings highlight the significant potential of SSM-based architectures in enabling high-accuracy and efficient brain MRI reconstruction.Item Open Access Epigraphnet: epilepsy recognition architecture with graph-based eeg analysis(Institute of Electrical and Electronics Engineers Inc., 2025) Şimşek, Ecem; Koç, Emirhan; Koç, AykutEarly and accurate detection of epileptic seizures is crucial for patients' quality of life and the treatment process. In this context, our study proposes EpiGrafNet, an innovative hybrid model that delivers high performance in epileptic seizure detection using electroencephalography (EEG) signals. The proposed method extracts both local and long-term temporal features from EEG signals via a one-dimensional convolutional neural network (CNN) and long short-term memory (LSTM) module; these features are then transformed into a graph structure by the utilization of correlation connection matrices (CCM), and integrated with graph convolutional neural networks (GCN). Experiments conducted with different sparsity values demonstrate that the developed model achieves superior accuracy, recall, precision, and F1 scores in both binary and multi-class epileptic seizure detection tasks, indicating that EpiGrafNet provides higher and more reliable results compared to existing methods. This approach, by effectively modeling the complex spatial and temporal relationships in EEG signals, significantly contributes to the development of automatic epileptic seizure detection systems in clinical applications.Item Open Access The ‘task’ of mind-wandering splits both multiple demand and default mode regions and ramps-up the deactivating regions(Elsevier, 2025-09-09) Giray, İrem; Farooqui, Ausaf AhmedThe activation of multiple demand (MD) regions to diverse tasks has been linked to the demands of making task-related cognitive control changes – keeping it focussed on task, controlling attention and working memory, organizing and maintaining a task model that will control the sequence and identity of what is to be done when, etc. Demanding tasks that require such control are also accompanied by a deliberative cognition whereby cognitive changes do not occur automatically and have to be made deliberately. We investigated whether the deliberativeness of cognition activates MD regions regardless of task-related demands. When not engaged in demanding tasks, the mind wanders. We asked participants to do the same during task periods, and to differentiate from rests, we asked them to deliberately and intensely wander their minds across random thoughts. We found that a set of MD regions – pre-supplementary motor area (preSMA), anterior insula, and posterior part of the middle frontal gyrus – activated during these periods, and another set – intraparietal sulcus, right anterior prefrontal cortex – deactivated. In fact, some of the activating regions (e.g., preSMA) activated more during this task than in response to robust working memory updating demands. Dissociations were also present in the Default Mode Network (DMN). Parts of the temporoparietal junction deactivated while posterior cingulate and medial prefrontal regions activated. Lastly, we found that the deactivating regions ramped-up their activity across the ‘task’ duration, showing that this ramp-up, previously linked to demands of sequentially organizing extended tasks, occurs during any construed task, including those without such demands.Item Open Access Correction to: 'Perceiving object size in pictures involves high-level processing' (2025), by Altan et al.(Royal Society Publishing, 2025-05-14) Altan, Ecem; Boyacı, Hüseyin; Dakin, Steven C.; Samuel Schwarzkopf D.The description under Population receptive field estimation (page 7 of PDF version) of how we calculated the noise ceiling of the fMRI data is inaccurate. Specifically, the sentence starting with Lastly, this measure was should instead read: This measure is equal to the noise ceiling, the maximum goodness of fit that can possibly be achieved for each vertex. Note that this is only an error in the text that does not affect the results. We want to correct this statement to avoid incorrect usage of noise ceiling calculations by others. We unreservedly apologize for this incorrect statement.Item Open Access Dynamic reorganization of functional networks underlying audiovisual interactions(Nature Research, 2025-11-12) Akdoğan, İrem; Aydin, Serap; Kafaligönül, HulusiCrossmodal interactions involve crosstalk between different cortical areas and dynamic recruitment of regions, which is crucial for integrating sensory information into a coherent percept. Despite their significance, the dynamic cortical networks underlying the crossmodal influence of auditory information on visual motion processing—particularly in terms of temporally resolved EEG connectivity—have yet to be comprehensively characterized. In the present study, we investigated frequency-specific networks underlying audiovisual interactions during motion and speed estimation. Functional networks were generated using directed transfer function (DTF) and adaptive DTF (ADTF) to estimate connectivity patterns of electroencephalogram (EEG) data. Network-based statistical analyses revealed frequency-specific networks in the theta and alpha bands, which supported long-range communication between occipital/parieto-occipital, parietal, and frontal regions during audiovisual interactions compared to unisensory visual motion processing. Graph theory analyses demonstrated a transition from localized and segregated processing to global integration, emphasizing cortical network reorganization according to the demands of sensory processing. Moreover, these analyses further revealed frequency-specific shifts in connectivity over time, with low-frequency oscillations exhibiting sustained connectivity increases, while high-frequency bands showed transient patterns, reflecting the temporal flexibility of neural networks. These findings illustrate how local and global network modulations reflect the brain’s dynamic reorganization, balancing integration and segregation during crossmodal influences.Item Open Access Effect of polygenic scores on the relationship between psychosis and cognition(Nature Publishing Group, 2025-11-21)Cognitive impairment is an important but often under-researched symptom in psychosis. Both psychosis and cognition are highly heritable and there is evidence of a genetic effect on the relationship between them. Using samples of adults (N = 4 506) and children (N = 10 981), we investigated the effect of schizophrenia and bipolar disorder polygenic scores on cognitive performance, and intelligence and educational attainment polygenic scores on psychosis presentation. Schizophrenia polygenic score was negatively associated with visuospatial processing in adults (beta: −0.0569; 95% confidence interval [CI]: −0.0926, −0.0212) and working memory (beta: −0.0432; 95% CI: −0.0697, −0.0168), processing speed (beta: −0.0491; 95% CI: −0.0760, −0.0223), episodic memory (betas: −0.0581 to −0.0430; 95% CIs: −0.0847, −0.0162), executive functioning (beta: −0.0423; 95% CI: −0.0692, −0.0155), fluid intelligence (beta: −0.0583; 95% CI: −0.0847, −0.0320), and total intelligence (beta: −0.0458; 95% CI: −0.0709, −0.0206) in children. Bipolar disorder polygenic score was not associated with any cognitive domains studied. Lower polygenic scores for intelligence were associated with greater odds of psychosis in adults (odds ratio [OR]: 0.886; 95% CI: 0.811–0.968). In children, lower polygenic scores for both intelligence (OR: 0.829; 95% CI: 0.777–0.884) and educational attainment (OR: 0.771; 95% CI: 0.724–0.821) were associated with greater odds of psychotic-like experiences. Our findings suggest that polygenic scores for both cognitive phenotypes and psychosis phenotypes are implicated in the relationship between psychosis and cognitive performance. Further research is needed to determine the direction of this effect and the mechanisms by which it occurs.Item Open Access Perceiving object size in pictures involves high-level processing(The Royal Society Publishing, 2025-11-26) Altan, Ecem; Boyacı, Hüseyin; Dakin, Steven C.; Schwarzkopf, D. SamuelThe description under Population receptive field estimation (page 7 of PDF version) of how we calculated the noise ceiling of the fMRI data is inaccurate. Specifically, the sentence starting with Lastly, this measure was should instead read: This measure is equal to the noise ceiling, the maximum goodness of fit that can possibly be achieved for each vertex. Note that this is only an error in the text that does not affect the results. We want to correct this statement to avoid incorrect usage of noise ceiling calculations by others. We unreservedly apologize for this incorrect statementItem Open Access GraphTeacher: transductive fine-tuning of encoders through graph neural networks(IEEE, 2025-10-31) Koç, Emirhan; Aras, Arda Can; Alikasifoglu, Tuna; Koç, AykutWe present GraphTeacher for fine-tuning transformer encoders by leveraging Graph Neural Networks (GNNs) to effectively train models when fully labeled training data is unavailable. When different percentages of labeled training data exist, we study popular transformer models, DistilBERT, RoBERTa, and BERT. The proposed approach uses the underlying graph structure of a corpus by allowing the transformer encoders to incorporate GNNs into the fine-tuning process. Using latent patterns and correlations identified in unlabeled data, our method aims to enhance the model’s adaptability to scarcely labeled data scenarios. Moreover, our approach excels in conducting single-instance inference, a capability not inherently possessed by models with a transductive (semi-supervised) training stage. GraphTeacher not only processes the unlabeled data effectively, as in transductive methods, but also offers an inductive inference setup for test samples. Experiments on diverse datasets and various GNN architectures show that integrating GNNs significantly enhances transformer encoders’ robustness and generalization capabilities, in particular under sparsely labeled training conditions. GraphTeacher demonstrates a noteworthy improvement, achieving up to a 10% increase in performance on the GLUE benchmark dataset compared to the baselines.Item Open Access An overview of medical image segmentation methods(Al-Nahrain University * College of Engineering, 2025-09-29) Jaber, Hussain A.; Al-Ghali, Basma A.; Kareem, Muna M.; Çankaya, Ilyas; Algın, OktayMedical image segmentation plays a crucial role in the realm of medical imaging. The process involves the division of an image to obtain a comprehensive view and ensure precise diagnostics. There are various methods that are employed, ranging from traditional approaches to the more advanced deep learning techniques. Both play a significant role in enhancing healthcare. With the continuous advancement in technology, there is a growing need for accurate segmentation. While traditional methods such as thresholding and region growing are effective, they may require human intervention for complex cases. Deep learning techniques, particularly Convolutional Neural Networks (CNNs), have significantly improved the process by learning intricate details and accurately segmenting the image. When these methods are combined, healthcare professionals can achieve high-quality, precise results. Furthermore, with the advancements in hardware and technology, real-time segmentation is now possible. Generally, the process of dividing medical images into segments is extremely important for the progress of healthcare with the help of artificial intelligence and the most recent advancements in the industry, such as explainable AI and multimodal learning. However, this meticulously detailed and in-depth review provides an all-encompassing and extensive analysis of the current methods utilized, their multitude of applications across various fields, and the promising emerging advancements that have the potential to pave the way for remarkable future improvements and innovations.Item Open Access Optimized single-channel head coil for maximizing drive field amplitude within safety limits(Infinite Science Publishing, 2025) Ozaslan, Ali Alper; Saritas, Emine Ulku; Top, Can BarışThis study investigates the design of an optimized single-channel head coil to maximize drive field (DF) amplitudes while minimizing the risk of peripheral nerve stimulation (PNS) in the human head. Using an optimization algorithm, we select the optimal winding positions on a fixed-length coil and achieve up to 25% increase in DF amplitude for superior portions of the head. Future work will analyze coil parameters and the choice of optimization constraints to increase DF amplitudes within the safety limits, across the entire human head.Item Open Access scGraPhT: Merging transformers and graph neural networks for single-cell annotation(Institute of Electrical and Electronics Engineers (IEEE), 2025-05-26) Koç, Emirhan; Kulkul, Emre; Kaynar, Gulara; Çukur, Tolga; Acar, Murat; Koç, AykutThe invention of single-cell RNA sequencing (scRNAseq) has enabled transcriptomic examination of cells on an individual basis, uncovering cell-to-cell phenotypic heterogeneity within isogenic cell populations. Inevitably, cell type annotation has emerged as a fundamental, albeit challenging task in scRNA-seq data analysis, which involves identifying and characterizing cells based on their unique molecular profiles. Recently, deep learning techniques with their data-driven priors have shown significant promise in this task. On the one hand, task-agnostic transformers pre-trained on large-scale biological databases capture generalizable representations but cannot characterize intricate relationships between genes and cells. Contrarily, task-specific graph neural networks (GNNs) trained on target datasets can characterize entity relationships, but they can suffer from poor generalizability. Furthermore, existing GNNs focus on either homogeneous or heterogeneous relationships, failing to capture the full cellular complexity. Here, we propose scGraPhT, a unified transformer–graph model that combines pre-trained transformer embeddings of scRNA-seq data with a multilayer GNN to capture cell-cell, cell-gene, and gene-gene relationships. Different from previous GNNs, scGraPhT examines both homogeneous and heterogeneous relationships through subgraph layers to offer a more comprehensive assessment. Since the graph construction uses transformer-derived embeddings, scGraPhT does not require costly training procedures and can also be adapted to leverage any transformer-based single-cell annotation method, such as scGPT or scBERT. Demonstrations on three scRNA-seq benchmark datasets indicate that scGraPhT outperforms state-of-the-art annotation methods without compromising efficiency. Utilizing Grad-CAM, we demonstrate how the GNN and transformer components complement each other to enhance performance. We share our source codes and datasets for reproducibility.Item Open Access Distraction by a human or a robot: Effects of perceptual load and action type(I E E E Computer Society, 2025-04-30) Özsu, Ataol Burak; Pekçetin, Tuǧçe Nur; Faydalı, Defne Şiir; Ürgen, Burcu AyşenThis study investigates how humans process and attend to robot actions compared to human actions under varying cognitive demands. Using a perceptual load paradigm, we examined whether robots capture attention similarly to humans and how this is influenced by the nature of their actions. Participants performed a letter detection task while being presented with task-irrelevant videos of either human or robot agents performing communicative or noncommunicative actions. Results demonstrated that both humans and robots captured attention through their actions, particularly when these actions were communicative. Under high perceptual load conditions, human distractors caused more interference than robot distractors, suggesting that agent identity becomes particularly important when cognitive resources are limited. These findings provide insights into how humans process robot actions in attentiondemanding situations and may have important implications for designing robot behaviors in operational contexts.Item Open Access Integrated MATLAB toolbox for fMRI visualization and data conversion(Iran University of Medical Sciences, 2025-07) Jaber, Hussain A.; Aljobouri, Hadeel Kassim; Koçak, Orhan Murat; Algın, Oktay; Çankaya, İlyasIntroduction: Working with functional magnetic resonance imaging (fMRI) often involves engaging with multiple file formats and complex viewers. In this study, we developed a novel platform as a visualization and conversion fMRI (VCfMRI) MATLAB toolbox for fMRI data. Methods: The VCfMRI was developed to read and write 3D fMRI volumes in DICOM, NIfTI, ANALYZE, and MAT formats and convert between them, on a single user-friendly platform. It includes 62 functions across seven graphical user interface modules for conversion, batch read/write, and orthogonal viewing (sagittal, coronal, horizontal). This toolbox also supports overlaying statistical maps on anatomical images with adjustable thresholds. We built and tested VCfMRI using real datasets from a scanner (3T, Siemens Co.) at UMRAM, Bilkent University. Results: VCfMRI successfully converted and visualized all supported formats in one environment, enabling synchronized 3D views and functional overlays with interactive threshold control, streamlining previously fragmented steps. Conclusion: The VCfMRI toolbox provides a simple and efficient solution for fMRI data conversion and visualization. It simplifies the handling of fMRI datasets across different formats, which is especially beneficial for physicians, healthcare specialists, and researchers who face challenges in processing and visualizing multi-format neuroimaging data.Item Open Access The impact of vitreous humor: a new perspective on radiation-induced cataractogenesis(Springer Science and Business Media Deutschland GmbH, 2025-09-23) Yiğit, E.; Koç, İ; Yazıcı, G.; Gümeler, E.; Elmalı, A.; Kahvecioğlu, A.; Yedekçi, F.Y.; Yabanoğlu Çiftçi, S.; Karataş, Meltem; Sezer, A.; Kıratlı, H.; Cengiz, M.Purpose: The precise mechanisms underlying radiation-induced cataractogenesis remain incompletely understood. Increased oxidative stress is known to play a central role in cataract pathogenesis. The vitreous humor contributes to maintaining the hypoxic environment of the lens by regulating oxygen pressure and containing antioxidants. This study aims to explore the effect of radiation-induced changes in the vitreous humor on lens health, with a particular focus on its cataractogenic potential. Methods: In this experimental study, 12 New Zealand rabbits were utilized. A single 20-Gy dose of radiation was administered to the left eye’s vitreous humor with a lens-sparing technique, while the right eye served as a control. Monthly ophthalmological evaluations were conducted over a 3-month period. At the end of the follow-up, orbital magnetic resonance imaging (MRI) was performed. Vitreous humor samples were analyzed using spectrophotometric methods to determine total oxidant and antioxidant levels. Results: Cataract formation was observed in two of the eight irradiated eyes (25%). The MRI analysis revealed a significant reduction in signal intensity within the left eye’s aqueous humor in non-contrast sequences (p = 0.03), while an increase in signal intensity was observed in late post-contrast sequences (p = 0.04). Spectrophotometric analysis indicated that total oxidant levels (p = 0.04) and the oxidative stress index (p = 0.04) were significantly lower in the treatment group. Conclusion: These findings suggest that radiation-induced changes in the vitreous humor and posterior ocular structures may influence the anterior chamber, contributing to the development of radiation-induced cataracts.Item Open Access Dc bias for improved baseline acquisition on a magnetic particle spectrometer setup(Infinite Science Publishing, 2025-03-14) Cantürk, Gülin; Kor, Ege; Sarıtaş, Emine ÜlküMagnetic particle spectrometer (MPS) setups typically feature a manually adjusted receive coil to minimize direct feedthrough interference. These adjustments can be compromised when a sample is physically inserted into the MPS setup or when a lengthy experiment is performed. In this work, we propose an MPS setup with a DC bias coil that can completely saturate the magnetic nanoparticle (MNP) signal, enabling baseline signal acquisition when the MNPs are inside the receive chamber.Item Open Access Editorial AI Reviewer (AIR) trial for responsible secure and efficient peer review(Institute of Electrical and Electronics Engineers, 2026-01-28) Wang, Ge; Çukur, Tolga; Kruger, Uwe; Ferina, Jennifer; Shan, HongmingItem Open Access Transcutaneous vagus Nerve stimulation enhances probabilistic learning(Wiley-Blackwell Publishing, Inc., 2025-03-09) Çakır, Resul; Büyükgüdük, İlkim; Bilim, Petek; Erdinç, Ataberk; Veldhuizen, Maria GeraldinetVNS enhances various memory and learning mechanisms, but there is inconclusive evidence on whether probabilistic learningcan be enhanced by tVNS. Here, we tested a simplified version of the probabilistic learning task with monetary rewards in abetween-participants design with left and right-sided cymba conchae and tragus stimulation (compared to sham stimulation)in a sample of healthy individuals (n = 80, 64 women, on average 26.38 years old). tVNS enhances overall accuracy significantly (p = 4.09 × 10⁻⁴) and reduces response times (p = 1.1006 × 10⁻⁴⁹) in the probabilistic learning phase. Reinforcement learningmodelling of the data revealed that the tVNS group uses a riskier strategy, dedicates more time to stimulus encoding and motorprocesses and exhibits greater reward sensitivity relative to the sham group. The learning advantage for tVNS relative to shampersists (p = 0.005 for accuracy and p = 9.2501 × 10⁻²⁷ for response times) during an immediate extinction phase with continuedstimulation in which feedback and reward were omitted. Our observations are in line with the proposal that tVNS enhances re-inforcement learning in healthy individuals. This suggests that tVNS may be useful in contexts where fast learning and learningpersistence in the absence of a reward is an advantage, for example, in the case of learning new habits.Item Open Access Visual appearance and sensitivity are mediated by distinct mechanisms(Elsevier Ltd, 2025-10-08) Karatok, Zahide Pamir; Boyacı, HüseyinIdentifying a visual stimulus and sensitivity to changes in its features have different requirements. Thus, it is possible that different mechanisms underlie appearance and sensitivity judgments of visual stimuli. Here, we tested this possibility using a complex scene where two patches with physically identical luminances appeared to have different lightness. Human participants first judged the perceived contrast of incremental and decremental gratings superimposed on the patches. Next, we measured detection thresholds. Finally, fMRI activity was recorded in response to gratings on these patches. We found that incremental, but not the decremental gratings, appeared to have higher contrast when superimposed on the perceptually lighter patch compared to the darker. However, the thresholds were lower for both types of gratings superimposed on the lighter patch compared to the darker. Finally, using fMRI, we found that the activity in the primary visual cortex (V1) aligns well with the results of the detection task. These results suggest that partly distinct mechanisms underlie sensitivity and appearance and, further, that V1 plays an important role in sensitivity judgments.Item Open Access scHyperLink revealing cell-type-specific gene regulation with hypergraph neural networks(Institute of Electrical and Electronics Engineers Inc., 2026-01-05) Kulkul, Emre; Çukur, Tolga; Koç, AykutSingle-cell RNA sequencing (scRNA-seq) allows gene expression to be measured at single-cell resolution, offering new opportunities to investigate Gene Regulatory Networks (GRNs), which represent the regulatory interactions between transcription factors (TFs) and their target genes. Given their relational structure, GRNs are formulated as graphs, enabling gene interaction inference to be framed as a link prediction task among graph nodes (i.e., genes). Prior work adopts Graph Neural Networks (GNNs) to this end, employing their unique ability to model inter-node relationships. However, since GNNs are inherently limited to pair-wise node interactions, they struggle to capture the higher-order dependencies characteristic of GRNs. Gene expression is regulated through multi-way feedback loops involving multiple TFs and targets, and disregarding these higher-order dependencies can lower accuracy in gene interaction inference. To overcome this limitation, we introduce scHyperLink, a hypergraph-based framework for GRN reconstruction. scHyperLink models gene interactions using Hypergraph Neural Networks (HGNNs), where hyperedges allow the simultaneous representation of multi-gene regulatory relationships. scHyperLink integrates experimentally derived interaction graphs with dynamically learned hyperedges to better reflect the underlying regulatory structure. We demonstrate that scHyperLink achieves higher accuracy than state-of-the-art on cell-type-specific benchmark datasets, particularly in sparse regimes with few known interactions. Moreover, we validate the biological relevance of scHyperLink via interpretability analyses on inferred hypergraphs and showcase its scalability to tissue-level analyses. We share the analyzed datasets and source codes for reproducibility.