WoS İndeksli Yayınlar Koleksiyonu
Permanent URI for this collectionhttps://hdl.handle.net/20.500.12573/394
Browse
12 results
Search Results
Article Frequency-Based Deep Occlusion Awareness Instance Segmentation(MDPI, 2026-02-26) Guzel, Yasin; Aydin, Zafer; Talu, Muhammed FatihOne major challenge faced by deep learning-based methods that detect target objects in the form of bounding boxes is object occlusion. High degrees of occlusion significantly diminish the accuracy of instance segmentation. Nonetheless, complex-valued Fourier descriptors can robustly represent object boundaries using minimal information. In this study, the impact of integrating Fourier descriptors-renowned for their strong representational capacity-with deep network models (UNet) that exhibit high generalization performance on instance segmentation accuracy was investigated. Within the scope of the research, nine network models were designed based on different strategies for utilizing frequency components. These variants fall into four strategy families: (i) UNet-style spectrum regression on fixed low-frequency windows (FUNet), (ii) magnitude-guided frequency selection/ROI construction (FUNet-Thr, FUNet-BBox), (iii) sequence models over tokenized FFT coefficients (BiLSTM Patch/Sorted), and (iv) encoder-only spectrum predictors with different depth/capacity (EncoderFFT1/2). To fairly evaluate the models' performance in segmenting objects subjected to disruptive factors (e.g., occlusion, blurring, noise), a specialized synthetic dataset was prepared. The task is formulated as single-target (single-instance), single-class segmentation. This dataset, automatically generated according to initial parameter values, contains images of objects moving at various speeds within a single frame. Among these models, the one termed FUNet, which relies on partial matching of central frequency components, achieved the highest segmentation accuracy despite the disruptive effects. Under the challenging Dataset 8 setting, the proposed FUNet achieved the highest overlap-based performance (Dice = 0.9329, IoU = 0.8842) among Attention U-Net, U-Net, and FourierNet, with statistically significant gains confirmed by paired per-image tests.Article BrAIn: A Comprehensive Artificial Intelligence-Based Morphology Analysis System for Brain Organoids and Neuroscience(Wiley, 2026-03-12) Polatli, Elifsu; Guner, Huseyin; Bastanlar, Yalin; Karakulah, Gokhan; Evranos, Ali Eren; Kahveci, Burak; Guven, SinanHuman-induced pluripotent stem cells (iPSCs) offer transformative potential for biomedical research, with iPSC-derived organoids providing more physiologically relevant models than traditional 2D cell cultures. Among these, brain organoids (BO) are particularly valuable for drug screening, disease modeling, and investigations into molecular pathways. Accurate representation of brain morphology is critical, as more complex organoid structures better mimic the human brain. Deep learning (DL) and machine learning (ML) approaches have become integral to analyzing organoid morphology, yet tools for comprehensive, time-resolved assessments are scarce. Here, we introduce BrAIn, a DL-based application for analyzing the developmental progression of BOs. BrAIn tracks their evolution from embryoid bodies (EBs) and quantifies parameters including area, Feret diameter, perimeter, roundness, and circularity. It also classifies budding and abnormal morphologies of 3D organoids and detects monolayer neural rosette structures, key features of neuronal differentiation. Designed with accessibility in mind, BrAIn provides a no-code interface, enabling researchers of all technical backgrounds to conduct advanced morphological analyses with ease. Our study demonstrates the application of BrAIn to evaluate the effects of different growth conditions-static, orbital shaker, and microfluidic chip-based-on BO development. Orbital shaker cultures resulted in the largest organoids, while chip-based systems achieved more homogeneous growth. Both conditions produced organoids with greater morphological complexity compared to static culture. BrAIn emerges as a robust, user-friendly tool to quantify BO development and explore how versatile growth conditions influence their morphology and maturation.Article Non-Contact Acoustic Screening for Sleep Apnea: A Subject-Aware Deep Learning Approach(Springer Science and Business Media Deutschland GmbH, 2026-02-11) Aygün Çakıroğlu, M.; Kizilkaya Aydoǧan, E.; Bolatturk, Ö.F.; Aydoğan, S.; Ismailoǧullari, S.; Delice, Y.Purpose: To explore the feasibility of using camera-derived, non-contact audio synchronized with PSG for clinically relevant sleep-apnea classification, and to benchmark compact deep models under a subject-aware design using a previously unstudied, real-world dataset. Methods: Thirty-two adults underwent simultaneous polysomnography (PSG) and camera-based non-contact audio recording. The synchronized audio segments were used to train and compare three compact deep-learning architectures (convolutional, attention-augmented, and transformer-based) under a subject-aware evaluation design that prevented identity leakage. Model performance and calibration were assessed at both segment and subject levels using standard statistical tests. Results: Subject-level evaluation was based on a very small, imbalanced test set of six subjects (one positive). Within this limited yet previously unstudied local dataset, the CNN_trans model achieved an apparent perfect ranking performance (AUC = 1.00; 95% CI 0.00–1.00), though this likely reflects the small, imbalanced test cohort, with recall = 1.00 and precision = 0.55. The wide confidence interval reflects substantial statistical uncertainty, and DeLong comparisons showed no significant AUC difference between CNN_trans and CNN_att (ΔAUC = − 0.042; p = 0.43). Conclusion: PSG-synchronized, non-contact audio supports accurate and well-calibrated sleep-apnea classification with compact deep models. This subject-aware evaluation suggests that contactless acoustic monitoring may have potential clinical relevance, motivating larger, multi-site validation. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2026.Article Deep-Learning Detection of Open-Apex Teeth on Panoramic Radiographs Using YOLO Models(Springer, 2025-12-23) Edik, Merve; Celebi, Fatma; Cukurluoglu, AykaganObjectivesThe use of deep learning in detecting teeth with open apices can prevent the need for additional radiographs for patients. The presented study aims to detect open-apex teeth using You Only Look Once (YOLO)-based deep learning models and compare these models.MethodsA total of 966 panoramic radiographs were included in the study. Open-apex teeth in panoramic radiographs were labeled. During the labeling process, they were divided into 6 classes in the maxilla and mandible, namely incisors, premolars, and molars. AI models YOLOv3, YOLOv4, and YOLOv5 were used. To evaluate the performance of the three detection models, both overall and separately for each class in the test dataset, precision, recall, average precision (mAP), and F1 score were calculated.ResultsYOLOv4 achieved the highest overall performance with a mean average precision (mAP) of 87.84% at IoU (Intersection over Union) 0.5 (mAP@0.5), followed by YOLOv5 with 85.6%, and YOLOv3 with 84.46%. Regarding recall, YOLOv4 also led with 90%, while both YOLOv3 and YOLOv5 reached 89%. Moreover, the F1 score was the highest for YOLOv4 (0.87), followed by YOLOv3 (0.86) and YOLOv5 (0.85).ConclusionsIn this study, YOLOv3, YOLOv4, and YOLOv5 were evaluated for the detection of open-apex teeth, and their mAP, recall, and F1 scores exceeded 84%. Deep learning-based systems can provide faster and more accurate results in the detection of open-apex teeth. This may help reduce the need for additional radiographs from patients and aid dentists by saving time.Conference Object Text Classification Experiments on Contextual Graphs Built by N-Gram Series(Springer International Publishing AG, 2025) Sen, Tarik Uveys; Yakit, Mehmet Can; Gumus, Mehmet Semih; Abar, Orhan; Bakal, GokhanTraditional n-gram textual features, commonly employed in conventional machine learning models, offer lower performance rates on high-volume datasets compared to modern deep learning algorithms, which have been intensively studied for the past decade. The main reason for this performance disparity is that deep learning approaches handle textual data through the word vector space representation by catching the contextually hidden information in a better way. Nonetheless, the potential of the n-gram feature set to reflect the context is open to further investigation. In this sense, creating graphs using discriminative ngram series with high classification power has never been fully exploited by researchers. Hence, the main goal of this study is to contribute to the classification power by including the long-range neighborhood relationships for each word in the word embedding representations. To achieve this goal, we transformed the textual data by employing n-gram series into a graph structure and then trained a graph convolution network model. Consequently, we obtained contextually enriched word embeddings and observed F1-score performance improvements from 0.78 to 0.80 when we integrated those convolution-based word embeddings into an LSTM model. This research contributes to improving classification capabilities by leveraging graph structures derived from discriminative n-gram series.Article Citation - WoS: 1Citation - Scopus: 1Spec17Tre: A New Dataset in Hardware Security and Using Deep Learning for Detecting Spectre Attacks(Springer Heidelberg, 2025-05-21) Aktas-Aydin, Hatice; Yalcin, GulayComputer performance has become a significant subject of study due to the processing of big data, the complexity of calculations and the importance of time efficiency. Many companies are improving processor operating principles to increase performance. The most common methods for this purpose are speculative execution and cache usage. While these techniques improve performance, they also introduce certain security vulnerabilities. Spectre is an attack that exploits vulnerabilities created by speculative execution, affecting all modern processor architectures. Research has shown that using machine learning to detect these attacks can be quite effective, although the features are typically gathered at the software level, which may limit detection since some performance parameters are not conveyed to the software. This study presents an analysis of Spectre attacks and their detection using machine learning and deep learning methods at the hardware level. Experiments are conducted using GEM5, a full-system hardware simulator, to ensure that only hardware-visible performance parameters are also collected. Attack detection is performed using Support Vector Machine (SVM) and Long Short-Term Memory (LSTM) methods. The LSTM method is used in conjunction with SVM and Convolutional Neural Network (CNN) techniques, and all models were tested on a new dataset, Spec17Tre, created using "519.lbm" from the SPEC CPU2017 benchmarks. The study achieved a 95% accuracy rate in attack detection using the LSTM + CNN hybrid model, which also yielded an F1 score of 0.999 for detecting applied Spectre attack scenarios.Article Citation - WoS: 1Citation - Scopus: 1Enhancing Diagnostic Quality in Panoramic Radiography: A Comparative Evaluation of GAN Models for Image Restoration(Wiley, 2025-09-04) Kolukisa, Burak; Celebi, Fatma; Ersu, Nihal; Yucel, Kemal Selcuk; Canger, Emin Murat; Murat Canger, EminPanoramic imaging is a widely utilized technique to capture a comprehensive view of the maxillary and mandibular dental arches and supporting facial structures. This study evaluates the potential of the Generative Adversarial Network (GAN) models-Pix2Pix, CycleGAN, and RegGAN-in enhancing diagnostic quality by addressing combinations of common image distortions. A panoramic radiograph data set was processed to simulate four types of distortions: (i) blurriness, (ii) noise, (iii) combined blurriness and noise, and (iv) anterior-region-specific blurriness. Three GAN models were trained and analyzed using quantitative metrics such as the peak signal-to-noise ratio (PSNR) and the structural similarity index measure (SSIM). In addition, two oral and maxillofacial radiologists conducted qualitative reviews to assess the diagnostic reliability of the generated images. Pix2Pix consistently outperformed CycleGAN and RegGAN, achieving the highest PSNR and SSIM values across all types of distortions. Expert evaluations also favored Pix2Pix, highlighting its ability to restore image accuracy and enhance clinical utility. CycleGAN showed moderate improvements in noise-affected images but struggled with combined distortions, while RegGAN yielded negligible enhancements. These findings underscore its potential for clinical application in refining radiographic imaging. Future research should focus on combining GAN techniques and utilizing larger datasets to develop universally robust image enhancement models.Article Citation - WoS: 2Citation - Scopus: 4Deep-Learning AI-Model for Predicting Dental Plaque in the Young Permanent Teeth of Children Aged 8-13 Years(MDPI, 2025-04-07) Tez, Banu Cicek; Guzel, Yasin; Eliacik, Bahar Basak Kiziltan; Aydin, Zafer; Kızıltan Eliaçık, Bahar BaşakBackground/Objectives: Dental plaque is a significant contributor to various prevalent oral health conditions, including caries, gingivitis, and periodontitis. Consequently, its detection and management are of paramount importance for maintaining oral health. Manual plaque assessment is time-consuming, error-prone, and particularly challenging in uncooperative pediatric patients. These limitations have encouraged researchers to seek faster, more reliable methods. Accordingly, this study aims to develop a deep learning model for detecting and segmenting plaque in young permanent teeth and to evaluate its diagnostic precision. Methods: The dataset comprises 506 dental images from 31 patients aged between 8 and 13 years. Six state-of-the-art models were trained and evaluated using this dataset. The U-Net Transformer model, which yielded the best performance, was further compared against three experienced pediatric dentists for clinical feasibility using 35 randomly selected images from the test set. The clinical trial was registered on under the ID NCT06603233 (1 June 2023). Results: The Intersection over Union (IoU) score of the U-Net Transformer on the test set was measured as 0.7845, and the p-values obtained from the three t-tests conducted for comparison with dentists were found to be below 0.05. Compared with three experienced pediatric dentists, the deep learning model exhibited clinically superior performance in the detection and segmentation of dental plaque in young permanent teeth. Conclusions: This finding highlights the potential of AI-driven technologies in enhancing the accuracy and reliability of dental plaque detection and segmentation in pediatric dentistry.Article Citation - WoS: 4Citation - Scopus: 7Combining N-Grams and Graph Convolution for Text Classification(Elsevier, 2025-05) Sen, Tarik Uveys; Yakit, Mehmet Can; Gumus, Mehmet Semih; Abar, Orhan; Bakal, GokhanText classification, a cornerstone of natural language processing (NLP), finds applications in diverse areas, from sentiment analysis to topic categorization. While deep learning models have recently dominated the field, traditional n-gram-driven approaches often struggle to achieve comparable performance, particularly on large datasets. This gap largely stems from deep learning' s superior ability to capture contextual information through word embeddings. This paper explores a novel approach to leverage the often-overlooked power of n-gram features for enriching word representations and boosting text classification accuracy. We propose a method that transforms textual data into graph structures, utilizing discriminative n-gram series to establish long-range relationships between words. By training a graph convolution network on these graphs, we derive contextually enhanced word embeddings that encapsulate dependencies extending beyond local contexts. Our experiments demonstrate that integrating these enriched embeddings into an long-short term memory (LSTM) model for text classification leads to around 2% improvements in classification performance across diverse datasets. This achievement highlights the synergy of combining traditional n-gram features with graph-based deep learning techniques for building more powerful text classifiers.Article Citation - WoS: 4Citation - Scopus: 5Beyond Visual Cues: Emotion Recognition in Images With Text-Aware Fusion(Elsevier, 2025-04) Sungur, Kerim Serdar; Bakal, GokhanSentiment analysis is a widely studied problem for understanding human emotions and potential outcomes. As it can be performed over textual data, working on visual data elements is also critically substantial to examining the current emotional status. In this effort, the aim is to investigate any potential enhancements in sentiment analysis predictions through visual instances by integrating textual data as additional knowledge reflecting the contextual information of the images. Thus, two separate models have been developed as image-processing and text-processing models in which both models were trained on distinct datasets comprising the same five human emotions. Following, the outputs of the individual models' last dense layers are combined to construct the hybrid multimodel empowered by visual and textual components. The fundamental focus is to evaluate the performance of the hybrid model in which the textual knowledge is concatenated with visual data. Essentially, the hybrid model achieved nearly a 3% F1-score improvement compared to the plain image classification model utilizing convolutional neural network architecture. In essence, this research underscores the potency of fusing textual context with visual information to refine sentiment analysis predictions. The findings not only emphasize the potential of a multi-modal approach but also spotlight a promising avenue for future advancements in emotion analysis and understanding.