Boran Sekeroglu

Plant recognition plays a vital role in agriculture and biodiversity monitoring, and deep learning, particularly convolutional neural networks (CNNs), has gained increased attention for automating this task. However, CNNs have a limitation in their ability to handle complex patterns due to the difficulty in capturing global contextual information. Furthermore, plant datasets are often created in laboratory environments that minimize discrimination challenges, enabling the analysis of model performance. This study proposes a hybrid deep learning model, HDL-PlantNet, for real-world plant recognition on the primary dataset, the Cyprus Seasonal Flora Image Dataset (CSFID), comprising 27 plant species. The HDL-PlantNet model integrates an EfficientNetV2-S convolutional backbone with a Transformer encoder to capture both spatial contextual and long-range dependencies. Additionally, the Swedish Leaf Dataset is used as a supplementary dataset to analyze the consistency of the HDL-PlantNet under controlled environments. Five benchmark CNN models are used for comparative evaluation, and statistical tests and an ablation study are conducted to assess the results. The proposed model achieved the highest observed Macro-F1 and Macro-AUC scores among the evaluated models, reaching 90.06% and 99.59%, respectively. The results demonstrate that combining convolutional and Transformer architectures yields computationally effective performance in fine-grained plant classification while maintaining a compact model size suitable for further research. This study contributes to real-time plant identification studies and supports informed ecological decision-making.

Objective: Febrile seizures (FS) are the most common seizures in childhood, yet identifying children at risk of developing epilepsy after the first FS remains challenging. We aimed to evaluate the prognostic potential of machine learning (ML) algorithms applied to post‐febrile seizure electroencephalography (EEG) recordings. Methods: We retrospectively reviewed 104 children (69 boys; mean age at first febrile seizure: 39.4 ± 18.2 months) who presented with their first febrile seizure between January 2018 and December 2021. Clinical data and EEG recordings obtained during N2 sleep were collected. EEG analysis was performed using separate preprocessing pipelines. For conventional EEG analysis, recordings were band‐pass filtered between 1 and 40 Hz, and artifact‐free segments were analyzed using Python‐based pipelines (YASA, MNE) to extract 34 time‐domain. The 34 extracted electrophysiological features were calculated across different bipolar EEG channels and evaluated together with aggregated inter‐channel measures, resulting in a total of 93 input attributes used for ML model development. High‐frequency oscillations (HFOs) were analyzed using a distinct pipeline applied to wideband EEG data before low‐pass filtering. Six machine learning algorithms—J48 Consolidated Decision Tree, Random Forest, Gradient Boosting, Extreme Gradient Boosting, k‐nearest neighbor, and Support Vector Machine—were trained using 10 × 7 repeated cross‐validation. Model performance was evaluated using sensitivity, specificity, accuracy, area under the receiver operating characteristic curve (ROC AUC), and F1‐score. Results: Over a median follow‐up of 4.2 months, 13 patients (12.5%) developed epilepsy, and all diagnoses were made within 9 months. XGBoost achieved the highest accuracy (0.89) and specificity (0.95) but had low sensitivity (0.46). J48 achieved the highest sensitivity (0.87) and ROC AUC (0.79), with a specificity of 0.71. Incorporating clinical features, including recurrent seizures, increased sensitivity to 0.95. The most relevant predictors were patient history, frequency band power, particularly increased power in lower frequency bands, and high‐frequency oscillations counts. Conclusion: ML‐based analysis of initial EEG after a first febrile seizure may assist in early epilepsy risk stratification. J48 provided superior sensitivity, and combining electroencephalography‐derived biomarkers with clinical data further enhanced predictive performance. Prospective, multicenter studies are warranted to confirm these findings.

Background: Celiac disease (CD) is a multisystem immune-mediated disorder increasingly recognized to affect sleep and neurobehavioral functioning. Pediatric data remain limited, and no prior study has examined especially for sleep microstructure in this population. This study evaluates the prevalence and patterns of sleep disturbances in children with CD using the Sleep Disturbance Scale for Children (SDSC) and explores potential electrophysiological correlates through N2 sleep spindle analysis. Methods: Children with biopsy-confirmed CD (n = 31) and age-matched controls (n = 25) completed the SDSC. A subgroup of CD patients with SDSC ≥ 35 and healthy controls underwent quantitative sleep spindle analysis (C3, C4, O1, O2) using automated and visual verification methods combined. Results: Clinically significant sleep disturbances were substantially more prevalent in CD than in controls (77.4% vs. 12%). Excessive somnolence, sleep–wake transition disorders, and sleep hyperhidrosis were the most affected domains. Moreover, among children with CD, those noncompliant with a gluten-free diet exhibited higher rates of excessive somnolence and sleep–wake transition disorders. While spindle parameters did not differ between groups, higher SDSC scores (≥35)—particularly in the somnolence and sleep–wake transition disorder domains—are associated with reduced spindle amplitude and density, suggesting that spindle alterations are linked to sleep disturbance severity rather than disease status per se. Conclusions: Sleep disturbances are common in pediatric CD and worsen with poor dietary adherence. Although sleep microarchitecture is largely preserved, reduced spindle activity is evident in children with higher subjective sleep burden, suggesting that spindle metrics may serve as potential objective markers for sleep disturbance. Longitudinal studies are required for validation.

Football is a complex game that requires combined technical, tactical, and psychological skills. The effect of training methods on players is crucial to improve their performance significantly. Different training methods can improve certain performance; however, the effect of combined training methodologies has not been sufficiently investigated. This study aimed to investigate the differential effects of small-sided games (SSGs), strength training (ST), and a combined training model (CTM) on the physical performance of soccer players. In this study, we analyzed 60 players in three groups. Two groups were trained independently with two single-training methods, small-sided games and a strength training model, and one group was trained independently with a double-training method, a combination of single-training methods. Before each training session, each group was given theoretical education specific to the training program they would perform. Eighteen physical measurements of the players were obtained using sensitive devices before and after they were completed. Four tree-based machine learning models, decision tree, random forest, gradient boosting, and extreme gradient boosting, were applied to solve the complex pattern of training strategies using the measurements. Extensive and comparative experiments were conducted to distinguish the groups of players. The distinction between the initial and final measurements was analyzed separately, and the extreme gradient boosting model achieved 0.73–0.80 and 1.00 accuracy for initial and final tests by outperforming other models. A superior model, XGBoost, was used to analyze the effective decisive factors that improved after the training sessions. The results showed that players in the double-training group demonstrated significantly greater improvements in skill performance compared to both single-training groups. In contrast, the skill gains observed in the single-training groups were modest and partially overlapping, indicating limited differentiation between them. These results suggest that integrated training programs may offer more comprehensive benefits and can inform evidence-based decision-making for coaches seeking to optimize physical development in soccer players.

Axial spondyloarthritis (SpA) is an inflammatory disease that causes back pain by affecting the axial skeleton, and the sacroiliac (SI) joints are mostly involved. An early diagnosis is required to prevent structural damage. Magnetic resonance imaging (MRI) is the primary tool for the diagnosis of axial spondyloarthritis; however, the occurrence of noninflammatory degenerative changes might prevent the accurate diagnosis even for experts. Deep learning (DL) aims to assist radiologists in detecting and diagnosing sacroiliitis by providing distinct and effective feature extraction along MR sequences. This retrospective study considers a primary dataset with 50 clinical sacroiliitis and 50 control group patients and aims to diagnose sacroiliitis using 4 MRI sequences. For this purpose, a simplified convolutional neural network model is developed, and comprehensive comparative experiments and analyses are performed. Three pre-trained DL models are considered in a comparative study using a transfer learning approach. Image-based, sequence-based, and patient-based experiments are conducted to evaluate general diagnostic abilities, analyze further clinical implementations, and determine the most informative and challenging MRI sequences. The results showed that the proposed model has the ability to detect sacroiliitis with 0.951 and 0.977 accuracy in patient and image-based experiments, respectively. The deep learning models obtained promising results for future clinical implementation to assist radiologists in detecting sacroiliitis. It is also analyzed that the STIR coronal images are the most challenging, while T1 axial sequences are the most informative sequences for sacroiliitis diagnosis.

High-frequency oscillations are oscillatory high-frequency (>80 Hz) signals that can be physiological or pathological. Physiological high-frequency oscillations related to sleep spindles have been reported. In addition, some reports indicate that high-frequency oscillations are a marker for the prediction of seizures. Our aim was to explore the presence of high-frequency oscillations in children with autism spectrum disorder and their relationship with clinical symptoms. Fifty-two children with autism spectrum disorder and 49 healthy children were enrolled in the study. EEG records were evaluated via the open-source Python library magnetoencephalography and electroencephalography (MNE)-high-frequency oscillation for the detection of scalp high-frequency oscillations and sleep spindles detected by open-source YASA. SPSS version 22 used for statistical analysis. The duration of sleep spindle was shorter, and the frequency was significantly greater, in children with autism spectrum disorder than in healthy controls. High-frequency oscillations unrelated to sleep spindle were more common in children with autism spectrum disorder (P < 0.01). Younger children with autism spectrum disorder had particularly more high-frequency oscillations unrelated to sleep spindle (P < 0.01). Within the autism spectrum disorder group of patients, high-frequency oscillations unrelated to sleep spindle were significantly greater in subgroups with moderate–severe social interaction problems (P < 0.01), moderate–severe restricted interest (P = 0.018), and severe linguistic delays (P = 0.006). High-frequency oscillations might be a candidate marker for autism spectrum disorder that might be used in clinical settings that warrants further studies.

(1) Background: Helminth infections are a widespread global health concern, with Ascaris and taeniasis representing two of the most prevalent infestations. Traditional diagnostic methods, such as egg-based microscopy, are fraught with challenges, including subjectivity and low throughput, often leading to misdiagnosis. This study evaluates the efficacy of advanced deep learning models in accurately classifying Ascaris lumbricoides and Taenia saginata eggs from microscopic images, proposing a technologically enhanced approach for diagnostics in clinical settings. (2) Methods: Three state-of-the-art deep learning models, ConvNeXt Tiny, EfficientNet V2 S, and MobileNet V3 S, are considered. A diverse dataset comprising images of Ascaris, Taenia, and uninfected eggs was utilized for training and validating these models by performing multiclass experiments. (3) Results: All models demonstrated high classificatory accuracy, with ConvNeXt Tiny achieving an F1-score of 98.6%, followed by EfficientNet V2 S at 97.5% and MobileNet V3 S at 98.2% in the experiments. These results prove the potential of deep learning in streamlining and improving the diagnostic process for helminthic infections. The application of deep learning models such as ConvNeXt Tiny, EfficientNet V2 S, and MobileNet V3 S shows promise for efficient and accurate helminth egg classification, potentially significantly enhancing the diagnostic workflow. (4) Conclusion: The study demonstrates the feasibility of leveraging advanced computational techniques in parasitology and points towards a future where rapid, objective, and reliable diagnostics are standard.

Enhancing image quality provides more interpretability for both human beings and machines. Traditional image enhancement techniques work well for specific uses, but they struggle with images taken in extreme conditions, such as varied distortions, noise, and contrast deformations. Deep-learning-based methods produce superior quality in enhancing images since they are capable of learning the spatial characteristics within the images. However, deeper models increase the computational costs and require additional modules for particular problems. In this paper, we propose self-supervised and supervised image enhancement models based on the time-shift image enhancement method (TS-IEM). We embedded the TS-IEM into a four-layer CNN model and reconstructed the reference images for the self-supervised model. The reconstructed images are also used in the supervised model as an additional layer to improve the learning process and obtain better-quality images. Comprehensive experiments and qualitative and quantitative analysis are performed using three benchmark datasets of different application domains. The results showed that the self-supervised model could provide reasonable results for the datasets without reference images. On the other hand, the supervised model outperformed the state-of-the-art methods in quantitative analysis by producing well-enhanced images for different tasks.