Scholarly Publications - UMRAM

Permanent URI for this collectionhttps://hdl.handle.net/11693/115674

Browse

Now showing 1 - 20 of 465

Access status: Open Access ,
A tutorial on MRI reconstruction: from modern methods to clinical implications
(IEEE Computer Society, 2025-10-03) Çukur, Tolga; Dar, Salman UH; Nezhad, Valiyeh A.; Jun, Yohan; Kim, Tae Hyung; Fujita, Shohei; Bilgiç, Berkin
MRI is an indispensable clinical tool, offering a rich variety of tissue contrasts to support broad diagnostic and research applications. Protocols can incorporate multiple structural, functional, diffusion, spectroscopic, or relaxometry sequences to provide complementary information for differential diagnosis, and to capture multidimensional insights into tissue structure and composition. However, these capabilities come at the cost of prolonged scan times, which reduce patient throughput, increase susceptibility to motion artifacts, and may require trade-offs in image quality or diagnostic scope. Over the last two decades, advances in image reconstruction algorithms-alongside improvements in hardware and pulse sequence design-have made it possible to accelerate acquisitions while preserving diagnostic quality. Central to this progress is the ability to incorporate prior information to regularize the solutions to the reconstruction problem. In this tutorial, we overview the basics of MRI reconstruction and highlight state-of-the-art approaches, beginning with classical methods that rely on explicit hand-crafted priors, and then turning to deep learning methods that leverage a combination of learned and crafted priors to further push the performance envelope. We also explore the translational aspects and eventual clinical implications of these methods. We conclude by discussing future directions to address remaining challenges in MRI reconstruction. The tutorial is accompanied by a Python toolbox (https://github.com/tutorial-MRI-recon/tutorial) to demonstrate select methods discussed in the article.
Access status: Open Access ,
Basal ganglia as an fMRI motor neurofeedback target in Parkinson’s disease
(Springer, 2025-11-19) Baqapuri, Halim I.; Terneusen, Anneke; Luehrs, Michael; Peters, Judith; Kuijf, Mark; Goebel, Rainer; Linden, David; Lozano, Andres M.; Mana, Josef; Jarraya, Bechir; Loução, Ricardo; Kocher, Martin; Visser-Vandewalle, Veerle; Çukur, Tolga
Parkinson’s disease (PD) is a progressive neurodegenerative disorder characterized by motor impairments. While pharmacological treatments offer symptom alleviation, their long-term effectiveness is insufficient. Deep Brain Stimulation (DBS) is a neurosurgical treatment that targets brain pathways to alleviate motor symptoms in PD. It is a highly invasive procedure and carries associated risks. This prompts investigation of non-invasive alternatives, such as real-time functional Magnetic Resonance Imaging (rt-fMRI) neurofeedback (NF). This work investigates the feasibility of using the basal ganglia, more specifically the putamen, a key structure in the motor network, as a potential NF target region. Two rt-fMRI studies were conducted: (i) Twelve healthy individuals participated in a single-blind, crossover study involving one MRI session targeting the putamen and the supplementary motor area (SMA) in separate runs. (ii) Twelve PD patients followed the same protocol but with three MRI sessions. We investigated whether participants could learn to voluntarily control brain activity through NF training. The PD patients successfully recruited the putamen during NF-reinforced motor imagery, which was also found at trend level in the healthy participants. We found no learning effect and no difference in putamen activation when it was directly targeted versus when the target signals came from the SMA. Overall, widespread cortical and subcortical areas involved in motor control were activated during neurofeedback. This study demonstrates for the first time that PD patients can modulate putamen activity through NF training, supporting its potential as a non-invasive neuromodulation target. This opens opportunities for integrating invasive and non-invasive neuromodulation for PD treatment.
Access status: Open Access ,
Top-down modulation of visual action perception: distinct task effects in the action observation network
(Springer, 2025-11-10) Eroğlu, Aslı; Ürgen, Burcu Ayşen
Perceiving others’ actions is essential for survival and social interaction. Cognitive neuroscience research has identified a network of brain regions crucial to visual action perception, known as the Action Observation Network (AON), comprising the posterior superior temporal cortex (pSTS), posterior parietal cortex, and premotor cortex. Recent research highlights the importance of integrating top-down processes, such as attention, to gain a deeper understanding of action perception. This study investigates how attention modulates the AON during human action perception. We conducted a two-session fMRI experiment with 27 participants. They viewed eight videos of pushing actions, varying in actor (female vs. male), effector (hand vs. foot), and target (human vs. object). In the first session, participants focused on specific features of the videos (actor, effector, or target). In the second, they passively viewed the videos. From the passive viewing session data, we defined regions of interest (ROIs) in the pSTS, parietal, and premotor cortices for each hemisphere. We then performed model-based representational similarity analysis (RSA) and decoding analysis. RSA results showed that only the task model, among all tested models, exhibited a significant correlation with neural representational similarity matrices (RDMs) across all ROIs, indicating a specific alignment between AON nodes and the ongoing task. Decoding analysis further showed that different task types uniquely affected each AON node, indicating feature- and region-specific interactions. These findings underscore that top-down attentional processes significantly alter neural representations within the AON, highlighting the dynamic interplay between attention and action perception in the brain.
Access status: Open Access ,
In subclinical depression in undergraduates, odor-evoked autobiographical memories are relatively less vivid than those evoked with words or photographs
(American Psychological Association, 2025-12) Yar, Bercem; Unguder, Yagmur; Veldhuizen, Maria Geraldine
he specificity of autobiographical memories is reduced in depression. Effects of depression on memory specificity have typically been examined using word cues, but odor cues may be more sensitive. Here, we measured depressive symptom severity in undergraduate students with and without subclinical depression (n = 99, 58 women, 40 men, and one nonbinary) and used word, photograph, or odor cues to examine memory retrieval. Overall, we observed more memories recalled in subclinical depression relative to no depression. We observed no effect of cue type or depression group on autobiographical memories specificity. However, subclinical depression was associated with lower memory pleasantness. Memory vividness was also associated with subclinical depression, mostly in memories evoked by odors. These results suggest that odors can be used to examine effects of depression on autobiographical memories. We highlight important aspects for future studies that can replicate these results with a more appropriate design and power.
Access status: Embargo ,
Formal thought disorder and familial risk in first-episode psychosis: a study of cortical thickness and neuroimaging-transcriptomic association analysis
(Elsevier, 2026-01-16) Çabuk, Tuğçe; Zhang, Y.; Palaniyappan, L.; Şahin Çevik, Didenur; Avcı, H.; Çakmak, I. B.; Yılmaz Kafalı, H.; Şenol, B.; Karlı Oğuz, K.; Toulopoulou, Timothea
Formal thought disorder (FTD), a prominent feature of schizophrenia, encompasses disruptions in thought, language, and communication. This study examines cortical thickness (CT) alterations in first-episode psychosis (FEP) patients (N = 24), their siblings (SIB) (N = 21), and healthy controls (CON) (N = 21) to explore potential neural correlates of FTD. Using structural MRI, we analyzed whole-brain CT and its relationship with positive and negative FTD measured by Thought and Language Index. Out-of-sample spatial correlations of gene expression with regional CT were also performed using a transcriptomic dataset. FEP had significant CT reductions in right middle frontal gyrus (MFG) compared with SIB and CON and in superior frontal gyrus (SFG) compared to CON; but SIB did not differ from CON. GLM analyses demonstrated that negative FTD exerted a significant main effect on CT in the MFG and SFG. By contrast, positive FTD showed no significant associations with CT. Neuroimaging-transcriptomic association analysis identified key biological pathways linked to cortical morphology. These findings emphasize the specific association between negative FTD and CT alterations in frontal brain regions, confirming prior reports. Future research should examine larger cohorts and investigate additional FTD subtypes to further elucidate neural correlates and potential familial risks of schizophrenia.
Access status: Open Access ,
Detection of microplastic waste by using a novel microfluidic system with an integrated object tracking algorithm
(Institute of Electrical and Electronics Engineers Inc., 2025-08-19) Khalak, Bushra Begum; Durmaz, Doruk; Külekçioğlu, Okan; Bahşi, Ela; Kasap, Selin; Onkal Engin, Güleda; Sarıtaş, Emine Ülkü; Erdem, Emine Yegan
Over the last couple of decades, microplastics, MPs, have become a large focus for pollution evaluation and sustainability. Water has become increasingly concentrated with MPs due to waste disposal and degradation over time. This project's goal is to detect and distinguish microplastics from other materials using a novelty microfluidic design. Analysis was done using multiple novel microchannel designs and Polydimethylsiloxane, PDMS, microchips. An object tracking algorithm, OT, was used to observe MPs and record flow through the microchannel. Footage tested on the microchips with the OT found that the varying pixel area shows the deformation of a microplastic under the chosen parameters.
Access status: Open Access ,
Bending dynamics of magnetic filaments at a curved bacterial bath interface
(American Chemical Society, 2025-12-12) Shafiei Aporvari, Mehdi; K. P. Velu, Sabareesh; Moradi, Ali-Reza; Sarıtaş, Emine Ülkü
The bending dynamics of microscopic filaments play a crucial role in various mechanical and biological processes. While thermal fluctuations typically have a minor effect, active fluctuations can provide sufficient mechanical energy to alter bending deformations in these systems. In this work, we investigate how active noise influences the bending dynamics of a self-assembled chain of magnetic particles at a curved liquid–air interface in the presence of swimming E. coli bacteria. We analyze the bending behavior of semiflexible chains under a lateral gravitational force and compare their response in passive and active baths. In a passive bath, the chain remains bent, fluctuating slightly around its deformed configuration due to thermal noise. However, in an active bath, strong bacterial activity suppresses the average bending deformation, causing the chain to fluctuate around a nearly straight configuration. These findings highlight the impact of active fluctuations on self-assembled microstructures and demonstrate a simple yet effective approach to studying the mechanical dynamics of microscopic filaments.
Access status: Open Access ,
Separating nanoparticle induced delays from relaxation time constant in TAURUS
(Infinite Science Publishing, 2025-03-14) Gülsün, Sevil Dilge; Alpman, Aslı; Sarıtaş, Emine Ülkü
Magnetic 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.
Access status: Open Access ,
Predictive processing in biological motion perception: evidence from human behavior
(Sage Publications Ltd., 2025-07-15) Elmas, Hüseyin Orkun; Er, Sena; Rezaki, Ada Dilek; İzgi, Ayşesu; Urgen, Buse M.; Boyacı, Hüseyin; Ürgen, Burcu Ayşen
Biological motion perception plays a crucial role in understanding the actions of other animals, facilitating effective social interactions. Although traditionally viewed as a bottom-up driven process, recent research suggests that top-down mechanisms, including attention and expectation, significantly influence biological motion perception at all levels, particularly highlighted under complex or ambiguous conditions. In this study, we investigated the effect of expectation on biological motion perception using a cued individuation task with point-light display (PLD) stimuli. We conducted three experiments investigating how prior information regarding action, emotion, and gender of PLD stimuli modulates perceptual processing. We observed a statistically significant congruency effect when preceding cues informed about action of the upcoming biological motion stimulus; participants performed slower in incongruent trials compared to congruent trials. This effect seems to be mainly driven from the 75% congruency condition compared to the non-informative 50% (chance level) validity condition. The congruency effect that was observed in the action experiment was absent in the emotion and gender experiments. These findings highlight the nuanced role of prior information in biological motion perception, particularly emphasizing that action-related cues, when moderately reliable, can influence biological motion perception. Our results are in line with the predictive processing framework, suggesting that the integration of top-down and bottom-up processes is context-dependent and influenced by the nature of prior information. Our results also emphasize the need to develop more comprehensive frameworks that incorporate naturalistic, complex and dynamic, stimuli to build better models of biological motion perception.
Access status: Open Access ,
Durum-uzay U-Net modeli ile öğrenme temelli MRG geriçatımı
(IEEE, 2025-08-15) Yavuz, Muhammet Talat; Öztürk, Şaban; Kabaş, Bilal; Çukur, Tolga
Manyetik Rezonans Görüntüleme (MRG), yumuşak dokuların yüksek kontrastlı görüntülenmesini sağlayan ve iyonlaştırıcı radyasyon içermeyen yaygın bir tıbbi görüntüleme yöntemidir. Ancak, uzun tarama süreleri klinik süreçlerde önemli kısıtlamalar oluşturmakta, bu da hasta konforunu azaltırken sağlık hizmetleri maliyetlerini artırmaktadır. Hızlandırılmış MRG teknikleri, tarama süresini kısaltmak amacıyla daha az frekans uzayı verisi toplamakta, ancak bu durum örnekleme eksikliklerinden kaynaklanan bozulmalara ve görüntü kalitesinin düşmesine yol açmaktadır. MRG geriçatım teknikleri, eksik örneklenen frekans uzayı verisinden yüksek kaliteli görüntüler üretmeyi hedefleyerek bu bozulmaları en aza indirmeyi amaçlamaktadır. Bu çalışmada, etkili bir MRG geriçatımı sağlamak amacıyla, düşük kaynak gereksinimiyle ardışık veri modelleme yeteneğine sahip ve uzun menzilli bağımlılıkları yakalayabilen ve yenilikçi bir Durum-Uzay Model tabanlı yöntem olan MambaCC’yi sunmaktayız. MambaCC, U-şekilli bir derin ögrenme ağı üzerinde çalışmakta olup, kanal karıştırma blokları ile desteklenerek görüntü bozulmalarının kanal bazındaki etkilerini de dikkate almaktadır. Yöntemin performansı fastMRI veri seti kullanılarak MoDL, U-Net, TransUNet ve UMamba gibi öncü yöntemlerle karşılaştırılmış olup, PSNR ve SSIM metrikleri açısından üstün performans gösterdiği belirlenmiştir. Elde edilen sonuçlar, Durum-Uzay tabanlı mimarilerin, yüksek doğruluklu ve verimli beyin MRG geriçatımı süreçlerinde önemli bir potansiyele sahip olduğunu ortaya koymaktadır.
Access status: Open Access ,
SSDiffusion: state-space diffusion model for medical image synthesis
(IEEE, 2025-08-15) Kabaş, Bilal; Nezhad, Valiyeh Ansarian; Atlı, Ömer Faruk; Arslan, Fuat; Çukur, Tolga
Medical image synthesis enables imputation of missing slices from acquired modalities, thereby reducing the need for repetitive and prolonged procedures in diagnostics. However, this is a non-trivial task due to the ill-posed and nonlinear characteristics of the problem. Generative adversarial networks (GANs) have shown promise in addressing these challenges. However, GANs can suffer from instability and mode collapse. Denoising diffusion models (DDMs) are recently demonstrated to be superior to GANs in terms of training stability and high fidelity. Yet, existing diffusion-based techniques rely on U-Net backbones, which are suboptimal for capturing contextual relationships between different tissue parts. Although transformers are adept at extracting the global context, their quadratic complexity makes them impractical for processing individual pixels rather than image patches. State-space models (SSMs) provide an efficient alternative with lower complexity, enabling the handling of extremely long sequences while effectively extracting contextual relationships. In this paper, we introduce SSDiffusion, a novel state-space diffusion model designed for multimodal medical image translation. By leveraging the strengths of state-space representations, our approach effectively captures contextual dependencies, improving the translation performance. We evaluate SSDiffusion on a variety of medical image translation tasks. Experiments indicate that SSDiffusion outperforms state-of-the-art GAN and diffusion methods.
Access status: Open Access ,
EpiGrafNet: grafik temelli EEG analizi ile epilepsi tanı mimarisi
(Institute of Electrical and Electronics Engineers Inc., 2025-08-15) Şimşek, Ecem; Koç, Emirhan; Koç, Aykut
Epileptik nöbetlerin erken ve doğru tespiti, hastaların yaşam kalitesi ve tedavi sürecinde hayati öneme sahiptir. Bu dogrultuda çalışmamız, elektroensefalografi (EEG) sinyalleri kullanarak epileptik nöbet tespitinde yüksek performans sunan yenilikçi bir hibrit model olan EpiGrafNet’i önermektedir. Önerilen yöntem, tek boyutlu konvolüsyonel sinir ağları (CNN) ve uzun-kısa süreli bellek (LSTM) modülü aracılığıyla EEG sinyallerinden yerel ve uzun vadeli zamansal öznitelikler çıkarmakta; bu öznitelikler, korelasyonel bağlantı matrisleri (KBM) kullanılarak grafik yapısına dönüştürülmekte ve grafik konvolüsyonel sinir ağları (GCN) ile entegre edilmektedir. Geliştirilen model, farklı seyreklik değerleriyle yapılan deneyler sonucunda hem ikili hem de çok sınıflı epileptik nöbet tespitinde üstün doğruluk, duyarlılık, kesinlik ve F1 skorları elde etmekte, bu da EpiGrafNet’in mevcut yöntemlere göre daha yüksek ve güvenilir sonuçlar verdiğini göstermektedir. Önerilen yaklaşım, EEG sinyallerindeki karmaşık uzamsal ve zamansal ilişkileri etkili bir şekilde modelleyerek, klinik uygulamalarda otomatik epileptik nöbet tespit sistemlerinin geliştirilmesine önemli katkılar sağlamaktadır.
Access status: 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 Ahmed
The 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.
Access status: 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.
Access status: Open Access ,
Dynamic reorganization of functional networks underlying audiovisual interactions
(Nature Research, 2025-11-12) Akdoğan, İrem; Aydin, Serap; Kafaligönül, Hulusi
Crossmodal 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.
Access status: Open Access ,
Effect of polygenic scores on the relationship between psychosis and cognition
(Nature Publishing Group, 2025-11-21) Varney, Lauren; Jedlovszky, Krisztina; Wang, Baihan; Murtough, Stephen; Cotic, Marius; Richards-Belle, Alvin; Saadullah Khani, Noushin; Lau, Robin; Abidoph, Rosemary; McQuillin, Andrew; Thygesen, Johan H.; Alizadeh, Behrooz Z.; Bender, Stephan; Crespo-Facorro, Benedicto; Hall, Jeremy; Iyegbe, Conrad; Kravariti, Eugenia; Lawrie, Stephen M.; Mata, Ignacio; McDonald, Colm; Murray, Robin M.; Prata, Diana; Toulopoulou, Timothea; van Haren, Neeltje EM; Bramon, Elvira
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.
Access status: Open Access ,
Emotion classification with visibility graphs
(Institute of Electrical and Electronics Engineers, 2025-05-05) Şimsek, Ecem; Topçu, Atakan; Koç, Emirhan; Sarıtaş, Emine Ülkü; Koç, Aykut
Transformers have gained prominence in natural language processing due to their representational capabilities and performances. Transformers process natural language as a sequence on finite context windows; however, global relationships among words beyond these windows cannot be completely modeled via sequence processing only. Graph neural network (GNN) based models have been proposed to alleviate this problem, as they provide geometric extensions to neural networks, enabling models to learn associations within a text. However, regular graph-based methods ignore the sequential nature of underlying texts. In this paper, we propose EmoVis, the first generic graph-based neural network that utilizes visibility graphs, which converts classical time-series information to graph representations. We cast the problem as an emotion classification task, enabling the proposed model to learn associations between the labels and words in a sentence. Moreover, EmoVis can be used as a highly modular graph-based extension to any transformer-based model, significantly improving their performance and learning capabilities in various languages. We experimentally show that EmoVis enables transformer-based models to outperform the state-of-the-art baselines across three diverse datasets in different languages in the SemEval2018 competition datasets and the GoEmotions dataset.
Access status: Open Access ,
GraphTeacher: transductive fine-tuning of encoders through graph neural networks
(IEEE, 2025-10-31) Koç, Emirhan; Aras, Arda Can; Alikasifoglu, Tuna; Koç, Aykut
We 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.
Access status: 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, Oktay
Medical 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.
Access status: Open Access ,
Optimized single-channel head coil for maximizing drive field amplitude within safety limits
(Infinite Science Publishing, 2025-03-14) Özaslan, Ali Alper; Sarıtaş, Emine Ülkü; Top, Can Barış
This 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.

Browse

Recent Submissions