Zeki Kuş
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Zeki Kuş has a total of 31 co-authors. Here are the top co-authors by number of shared publications:
- Ayla Gülcü
Bahcesehir University
6 shared publications - Berna Kiraz
Fatih Sultan Mehmet Vakif University
6 shared publications - Alper Kiraz
Koc University
5 shared publications - Musa Aydin
Computer Engineering Department, Fatih Sultan Mehmet Vakıf University, İstanbul, Türkiye
5 shared publications - Berna Kiraz
AI and Data Engineering, Istanbul Technical University, İstanbul, Türkiye
3 shared publications - Musa Aydin
Department of Artificial Intelligence and Data Engineering, Samsun University, Samsun, Türkiye
3 shared publications
Latest Publications
- Open Access
- 6 May 2026
- Published by Springer Nature in Multimedia Systems
- Vol. 32 (4) , 245
Neural architecture search for 3D biomedical image classification
3D medical image classification is crucial for improving diagnostic accuracy and treatment planning, but it encounters challenges due to the complexity and variability of volumetric data. While 3D Convolutional Neural Networks offer potential solutions, designing effective architectures is complex and resource-intensive. Neural Architecture Search automates this process, optimizing network designs for specific tasks, thereby improving model performance. This study introduces a novel extension of the PBC-NAS method for 3D medical image classification, aiming to balance prediction accuracy and model complexity. We focus on optimizing neural network architectures using Neural Architecture Search for six different 3D datasets from MedMNIST3D, including OrganMNIST3D, NoduleMNIST3D, FractureMNIST3D, AdrenalMNIST3D, VesselMNIST3D, and SynapseMNIST3D, which are derived from real-world clinical imaging datasets. We have compared our method with state-of-the-art handcrafted networks, AutoML frameworks and recent NAS studies in terms of prediction performance and model complexity. The proposed NAS methods demonstrate superior performance compared to state-of-the-art handcrafted networks and AutoML frameworks. Our proposed model (Ours #3 $$^\dagger$$ ) achieves the highest average Area Under the Curve (AUC) of 0.915 and accuracy (ACC) of 0.847 (best result across three independent runs), outperforming all handcrafted networks and AutoML frameworks. Compared to other NAS-based methods, all proposed models achieve higher average AUC scores, and it is important to note that they do not rely on data augmentation, pre-processing, or feature selection, unlike the competing NAS methods which do use data augmentation during training. The study also highlights significant reductions in computational complexity, with FLOPs reduced by up to 45.51 times and parameters by up to 211 times compared to ResNet models. An ablation study reveals that while fine-tuning a model optimized for one dataset can achieve competitive results on other datasets, dataset-specific NAS is crucial for optimal performance. Despite this, the ablation results still outperform ResNets and AutoML frameworks in terms of average AUC and ACC. The study concludes that the proposed NAS approach effectively optimizes neural network architectures for complex 3D medical image classification tasks, achieving state-of-the-art performance without data augmentation.- 6 May 2026
- Published by Springer Nature in Multimedia Systems
- Vol. 32 (4) , 245
Open AccessNeural architecture search for 3D biomedical image classification
3D medical image classification is crucial for improving diagnostic accuracy and treatment planning, but it encounters challenges due to the complexity and variability of volumetric data. While 3D Convolutional Neural Networks offer potential solutions, designing effective architectures is complex and resource-intensive. Neural Architecture Search automates this process, optimizing network designs for specific tasks, thereby improving model performance. This study introduces a novel extension of the PBC-NAS method for 3D medical image classification, aiming to balance prediction accuracy and model complexity. We focus on optimizing neural network architectures using Neural Architecture Search for six different 3D datasets from MedMNIST3D, including OrganMNIST3D, NoduleMNIST3D, FractureMNIST3D, AdrenalMNIST3D, VesselMNIST3D, and SynapseMNIST3D, which are derived from real-world clinical imaging datasets. We have compared our method with state-of-the-art handcrafted networks, AutoML frameworks and recent NAS studies in terms of prediction performance and model complexity. The proposed NAS methods demonstrate superior performance compared to state-of-the-art handcrafted networks and AutoML frameworks. Our proposed model (Ours #3 $$^\dagger$$ ) achieves the highest average Area Under the Curve (AUC) of 0.915 and accuracy (ACC) of 0.847 (best result across three independent runs), outperforming all handcrafted networks and AutoML frameworks. Compared to other NAS-based methods, all proposed models achieve higher average AUC scores, and it is important to note that they do not rely on data augmentation, pre-processing, or feature selection, unlike the competing NAS methods which do use data augmentation during training. The study also highlights significant reductions in computational complexity, with FLOPs reduced by up to 45.51 times and parameters by up to 211 times compared to ResNet models. An ablation study reveals that while fine-tuning a model optimized for one dataset can achieve competitive results on other datasets, dataset-specific NAS is crucial for optimal performance. Despite this, the ablation results still outperform ResNets and AutoML frameworks in terms of average AUC and ACC. The study concludes that the proposed NAS approach effectively optimizes neural network architectures for complex 3D medical image classification tasks, achieving state-of-the-art performance without data augmentation. - Open Access
- 6 February 2026
- Published by Springer Nature in Scientific Data
- Vol. 13 (1)
A Large-Scale Peripheral Blood Cell Dataset for Automated Hematological Analysis
White blood cell classification is fundamental to hematological diagnosis, yet existing datasets are limited in scale and class diversity. We present a comprehensive peripheral blood cell dataset comprising 31,489 high-resolution microscopic images across 13 distinct cell classes, representing the largest publicly available collection for automated blood cell analysis. Images are acquired using the Sysmex DI-60 system from May-Grünwald-Giemsa-stained blood smears at 100 × magnification under standardized laboratory conditions. Expert hematologists with over 10 years of experience performed manual annotation with high inter-rater agreement (Cohen’s kappa >0.85 for all classes). The dataset includes common cell types such as segmented neutrophils and lymphocytes, alongside diagnostically critical but rare subtypes, including myelocytes, blasts, and reactive lymphocytes. Images are organized into training, validation, and test splits (70:10:20 ratio) with consistent 368 × 368 pixel resolution. Baseline experiments using 14 deep learning architectures demonstrate the dataset’s utility, with DenseNet-121 achieving 95.23% accuracy. KU-Optofil PBC Dataset addresses critical gaps in medical image analysis datasets and supports the development of robust automated hematology systems for clinical applications.
- 6 February 2026
- Published by Springer Nature in Scientific Data
- Vol. 13 (1)
Open AccessA Large-Scale Peripheral Blood Cell Dataset for Automated Hematological Analysis
White blood cell classification is fundamental to hematological diagnosis, yet existing datasets are limited in scale and class diversity. We present a comprehensive peripheral blood cell dataset comprising 31,489 high-resolution microscopic images across 13 distinct cell classes, representing the largest publicly available collection for automated blood cell analysis. Images are acquired using the Sysmex DI-60 system from May-Grünwald-Giemsa-stained blood smears at 100 × magnification under standardized laboratory conditions. Expert hematologists with over 10 years of experience performed manual annotation with high inter-rater agreement (Cohen’s kappa >0.85 for all classes). The dataset includes common cell types such as segmented neutrophils and lymphocytes, alongside diagnostically critical but rare subtypes, including myelocytes, blasts, and reactive lymphocytes. Images are organized into training, validation, and test splits (70:10:20 ratio) with consistent 368 × 368 pixel resolution. Baseline experiments using 14 deep learning architectures demonstrate the dataset’s utility, with DenseNet-121 achieving 95.23% accuracy. KU-Optofil PBC Dataset addresses critical gaps in medical image analysis datasets and supports the development of robust automated hematology systems for clinical applications.
- 3 February 2026
- Published by Springer Nature in Progress in Artificial Intelligence
- Vol. 15 (3) , 707-721
TripletMAML: A metric-based model-agnostic meta-learning algorithm for few-shot classification
- 3 February 2026
- Published by Springer Nature in Progress in Artificial Intelligence
- Vol. 15 (3) , 707-721
TripletMAML: A metric-based model-agnostic meta-learning algorithm for few-shot classification
- 16 November 2025
- Published by Springer Nature in Health Information Science and Systems
- Vol. 13 (1) , 78
MOLiNAS: multi-objective lightweight neural architecture search for whole-slide multi-class blood cell segmentation
Blood cell analysis plays a key role in clinical diagnosis and hematological research. The accurate identification and quantification of different blood cell types is essential for the diagnosis of various diseases. The conventional manual method of blood cell analysis is both laborious and time-consuming, highlighting the need for automated segmentation techniques. In this paper, the blood cell segmentation problem is considered as a multi-class segmentation problem to detect the different types of blood cells in a given image. Two new multi-objective lightweight neural architecture search (NAS) algorithms (MOLiNAS) are designed to tackle the challenge of whole-slide multi-class blood cell segmentation problems. Our approaches integrate the most advantageous aspects of different approaches to search for the best U-shaped network architecture. The performance of our approaches is compared with lightweight networks and NAS studies in the literature. Our best solution (MOLiNASv2_sol3) achieves an IoU of 87.33 ± 1.53%, F1 score of 91.69 ± 1.20%, Precision of 93.50 ± 1.15%, and Recall of 91.34 ± 0.01%, outperforming lightweight networks such as EfficientNet, MobileNetv2, and MobileNetv3 across all segmentation metrics. Moreover, our approaches demonstrate highly competitive performance by utilizing up to 7.38 times fewer FLOPs and up to 4.03 times fewer trainable parameters than existing NAS studies while requiring only 0.07 million parameters. Additionally, ablation studies and cross-dataset evaluations demonstrate the robustness and generalizability of our approach.
- 16 November 2025
- Published by Springer Nature in Health Information Science and Systems
- Vol. 13 (1) , 78
MOLiNAS: multi-objective lightweight neural architecture search for whole-slide multi-class blood cell segmentation
Blood cell analysis plays a key role in clinical diagnosis and hematological research. The accurate identification and quantification of different blood cell types is essential for the diagnosis of various diseases. The conventional manual method of blood cell analysis is both laborious and time-consuming, highlighting the need for automated segmentation techniques. In this paper, the blood cell segmentation problem is considered as a multi-class segmentation problem to detect the different types of blood cells in a given image. Two new multi-objective lightweight neural architecture search (NAS) algorithms (MOLiNAS) are designed to tackle the challenge of whole-slide multi-class blood cell segmentation problems. Our approaches integrate the most advantageous aspects of different approaches to search for the best U-shaped network architecture. The performance of our approaches is compared with lightweight networks and NAS studies in the literature. Our best solution (MOLiNASv2_sol3) achieves an IoU of 87.33 ± 1.53%, F1 score of 91.69 ± 1.20%, Precision of 93.50 ± 1.15%, and Recall of 91.34 ± 0.01%, outperforming lightweight networks such as EfficientNet, MobileNetv2, and MobileNetv3 across all segmentation metrics. Moreover, our approaches demonstrate highly competitive performance by utilizing up to 7.38 times fewer FLOPs and up to 4.03 times fewer trainable parameters than existing NAS studies while requiring only 0.07 million parameters. Additionally, ablation studies and cross-dataset evaluations demonstrate the robustness and generalizability of our approach.
- Open Access
- 19 June 2025
- Published by Springer Nature in Health Information Science and Systems
- Vol. 13 (1) , 1-18
Multi-teacher based knowledge distillation for retinal vessel segmentation
Accurate segmentation of retinal vessels is crucial for the early diagnosis and management of various ocular diseases. Existing methods often struggle to segment thin vessels, leading to missed diagnoses and inaccurate treatment plans. This study proposes a novel Multi-Teacher Based Knowledge Distillation (MTKD) method for Retinal Vessel Segmentation (RVS) to address this challenge. Our approach utilizes the expertise of multiple teacher networks, each specialized in learning different vessel characteristics. Specifically, we train three distinct teacher networks: one on the original ground truth, one on a modified ground truth highlighting thin vessels, and another on a modified ground truth emphasizing thick vessels. The student network is then trained to minimize the knowledge discrepancy between its predictions and the soft predictions of all three teachers. By incorporating knowledge from these specialized teachers, the student network effectively learns to segment both thin and thick vessels with improved accuracy. We evaluate our method on two retinal fundus image datasets and two angiography datasets, demonstrating highly competitive performance compared to state-of-the-art methods. The proposed method improves the baseline U-Net model by up to 8.44 points in F1 and 10.42 points in IOU. Additionally, we introduce a penalization technique to the student model’s loss function, further enhancing segmentation performance. Comprehensive ablation studies validate the effectiveness of the multi-teacher approach, the choice of loss functions, and the impact of model complexity. Our findings suggest that MTKD offers a promising approach for enhancing the robustness and accuracy of RVS. All source code, datasets, and results are made publicly available to support reproducibility and further research.
- 19 June 2025
- Published by Springer Nature in Health Information Science and Systems
- Vol. 13 (1) , 1-18
Open AccessMulti-teacher based knowledge distillation for retinal vessel segmentation
Accurate segmentation of retinal vessels is crucial for the early diagnosis and management of various ocular diseases. Existing methods often struggle to segment thin vessels, leading to missed diagnoses and inaccurate treatment plans. This study proposes a novel Multi-Teacher Based Knowledge Distillation (MTKD) method for Retinal Vessel Segmentation (RVS) to address this challenge. Our approach utilizes the expertise of multiple teacher networks, each specialized in learning different vessel characteristics. Specifically, we train three distinct teacher networks: one on the original ground truth, one on a modified ground truth highlighting thin vessels, and another on a modified ground truth emphasizing thick vessels. The student network is then trained to minimize the knowledge discrepancy between its predictions and the soft predictions of all three teachers. By incorporating knowledge from these specialized teachers, the student network effectively learns to segment both thin and thick vessels with improved accuracy. We evaluate our method on two retinal fundus image datasets and two angiography datasets, demonstrating highly competitive performance compared to state-of-the-art methods. The proposed method improves the baseline U-Net model by up to 8.44 points in F1 and 10.42 points in IOU. Additionally, we introduce a penalization technique to the student model’s loss function, further enhancing segmentation performance. Comprehensive ablation studies validate the effectiveness of the multi-teacher approach, the choice of loss functions, and the impact of model complexity. Our findings suggest that MTKD offers a promising approach for enhancing the robustness and accuracy of RVS. All source code, datasets, and results are made publicly available to support reproducibility and further research.
- 5 January 2025
- Published by Elsevier in Artificial Intelligence in Medicine
- Vol. 160, 103064
Neural Architecture Search for biomedical image classification: A comparative study across data modalities
- 5 January 2025
- Published by Elsevier in Artificial Intelligence in Medicine
- Vol. 160, 103064
Neural Architecture Search for biomedical image classification: A comparative study across data modalities
- 30 December 2024
- Published by Gazi University in Gazi University Journal of Science Part A: Engineering and Innovation
- Vol. 11 (4) , 742-758
UKnow-Net: Knowledge-Enhanced U-Net for Improved Retinal Vessel Segmentation
Retinal vessel segmentation plays a critical role in diagnosing and managing ophthalmic and systemic diseases, as abnormalities in retinal vasculature can indicate disease progression. Traditional manual segmentation by expert ophthalmologists is time-consuming, labor-intensive, and prone to variability, underscoring the need for automated methods. While deep learning approaches like U-Net have advanced retinal vessel segmentation, they often struggle to generalize across diverse datasets due to differences in image acquisition techniques, resolutions, and patient demographics. To address these challenges, I propose UKnow-Net, a knowledge-enhanced U-Net architecture designed to improve retinal vessel segmentation across multiple datasets. UKnow-Net employs a multi-step process involving knowledge distillation and enhancement techniques. First, I train four specialized teacher networks separately on four publicly available retinal vessel segmentation datasets—DRIVE, CHASE_DB1, DCA1, and CHUAC—allowing each to specialize in the unique features of its respective dataset. These teacher networks generate pseudo-labels representing their domain-specific knowledge. We then train a student network using the ensemble of pseudo-labels from all teacher networks, effectively distilling the collective expertise into a unified model capable of generalizing across different datasets. Experiments demonstrate that UKnow-Net outperforms traditional handcrafted networks (such as U-Net, UNet++, and Attention U-Net) and several state-of-the-art models in key performance metrics, including sensitivity, specificity, F1 score, and Intersection over Union (IoU). Specifically, our two variants, UKnowNet-A and UKnowNet-B, show well performance; UKnowNet-A, trained solely on pseudo-labels, achieved higher sensitivity across all datasets, indicating a superior ability to detect true positives, while UKnowNet-B, which combines pseudo-labels with ground truth annotations, achieved balanced precision and recall, leading to higher F1 scores and IoU metrics. The integration of pseudo-labels effectively transfers the collective expertise of the teacher networks to the student network, enhancing generalization and robustness. I aim to ensure fair comparison and reproducibility in future research by publicly sharing our source code and model weights.
- 30 December 2024
- Published by Gazi University in Gazi University Journal of Science Part A: Engineering and Innovation
- Vol. 11 (4) , 742-758
UKnow-Net: Knowledge-Enhanced U-Net for Improved Retinal Vessel Segmentation
Retinal vessel segmentation plays a critical role in diagnosing and managing ophthalmic and systemic diseases, as abnormalities in retinal vasculature can indicate disease progression. Traditional manual segmentation by expert ophthalmologists is time-consuming, labor-intensive, and prone to variability, underscoring the need for automated methods. While deep learning approaches like U-Net have advanced retinal vessel segmentation, they often struggle to generalize across diverse datasets due to differences in image acquisition techniques, resolutions, and patient demographics. To address these challenges, I propose UKnow-Net, a knowledge-enhanced U-Net architecture designed to improve retinal vessel segmentation across multiple datasets. UKnow-Net employs a multi-step process involving knowledge distillation and enhancement techniques. First, I train four specialized teacher networks separately on four publicly available retinal vessel segmentation datasets—DRIVE, CHASE_DB1, DCA1, and CHUAC—allowing each to specialize in the unique features of its respective dataset. These teacher networks generate pseudo-labels representing their domain-specific knowledge. We then train a student network using the ensemble of pseudo-labels from all teacher networks, effectively distilling the collective expertise into a unified model capable of generalizing across different datasets. Experiments demonstrate that UKnow-Net outperforms traditional handcrafted networks (such as U-Net, UNet++, and Attention U-Net) and several state-of-the-art models in key performance metrics, including sensitivity, specificity, F1 score, and Intersection over Union (IoU). Specifically, our two variants, UKnowNet-A and UKnowNet-B, show well performance; UKnowNet-A, trained solely on pseudo-labels, achieved higher sensitivity across all datasets, indicating a superior ability to detect true positives, while UKnowNet-B, which combines pseudo-labels with ground truth annotations, achieved balanced precision and recall, leading to higher F1 scores and IoU metrics. The integration of pseudo-labels effectively transfers the collective expertise of the teacher networks to the student network, enhancing generalization and robustness. I aim to ensure fair comparison and reproducibility in future research by publicly sharing our source code and model weights.
- Open Access
- 25 November 2024
- Published by Springer Nature in Scientific Data
- Vol. 11 (1) , 1-15
MedSegBench: A comprehensive benchmark for medical image segmentation in diverse data modalities
MedSegBench is a comprehensive benchmark designed to evaluate deep learning models for medical image segmentation across a wide range of modalities. It covers a wide range of modalities, including 35 datasets with over 60,000 images from ultrasound, MRI, and X-ray. The benchmark addresses challenges in medical imaging by providing standardized datasets with train/validation/test splits, considering variability in image quality and dataset imbalances. The benchmark supports binary and multi-class segmentation tasks with up to 19 classes and uses the U-Net architecture with various encoder/decoder networks such as ResNets, EfficientNet, and DenseNet for evaluations. MedSegBench is a valuable resource for developing robust and flexible segmentation algorithms and allows for fair comparisons across different models, promoting the development of universal models for medical tasks. It is the most comprehensive study among medical segmentation datasets. The datasets and source code are publicly available, encouraging further research and development in medical image analysis.
- 25 November 2024
- Published by Springer Nature in Scientific Data
- Vol. 11 (1) , 1-15
Open AccessMedSegBench: A comprehensive benchmark for medical image segmentation in diverse data modalities
MedSegBench is a comprehensive benchmark designed to evaluate deep learning models for medical image segmentation across a wide range of modalities. It covers a wide range of modalities, including 35 datasets with over 60,000 images from ultrasound, MRI, and X-ray. The benchmark addresses challenges in medical imaging by providing standardized datasets with train/validation/test splits, considering variability in image quality and dataset imbalances. The benchmark supports binary and multi-class segmentation tasks with up to 19 classes and uses the U-Net architecture with various encoder/decoder networks such as ResNets, EfficientNet, and DenseNet for evaluations. MedSegBench is a valuable resource for developing robust and flexible segmentation algorithms and allows for fair comparisons across different models, promoting the development of universal models for medical tasks. It is the most comprehensive study among medical segmentation datasets. The datasets and source code are publicly available, encouraging further research and development in medical image analysis.