Yazılım Mühendisliği Bölümü

Bu koleksiyon için kalıcı URI

https://hdl.handle.net/20.500.13055/51

Listeleniyor 1 - 20 / 36

Multi-class network intrusion detection: Machine and deep learning benchmark with live stream deployment
(IEEE, 2026) Danışmaz, Betül; Bıyık, Mustafa Emre; Gözüaçık, Necip
The rapid rise of advanced cyber threats is exposing the limitations of traditional, signature-based network intrusion detection systems. While machine learning and deep learning models offer proactive defense mechanisms, current research is often compromised due to temporal data leakage, impractical threat classifications, and significant discrepancies between offline evaluation and online deployment. This study presents a comparative evaluation of traditional machine learning and deep learning architectures integrated into a fully functional real-time inference pipeline. Additionally, we propose an operationally motivated three-class classification encompassing Benign, Volumetric, and Semantic traffic to contextualize incident response. When the trained XGBoost, Random Forest, Decision Tree, LSTM, and BiLSTM models were evaluated and compared under the same constraints, the BiLSTM model achieved the highest Macro F1 score at 97.72%. However, XGBoost closely followed with a Macro F1 of 95.87%, while achieving 15.5 times faster inference. To translate these findings into practice, a Kafka-based live system architecture was designed that integrates real-time stream inference, zero-downtime model switching, and automated threat response capabilities.
Driver fatigue estimation using multimodal deep learning with eeg and eog signals
(IEEE, 2026) Demir, Șevval; Yapar, Rümeysa; Aldaş, Ahsen Nur; Gözüaçık, Necip
Driver fatigue is a critical condition that increases accident risk by causing loss of concentration and perceptual limitations. Therefore, researchers have focused on developing systems aimed at the early detection of driver fatigue. In this study, driver fatigue detection was targeted using Differential Entropy–Linear Dynamic System (DE-LDS) based Electroencephalography (EEG) features and Electrooculography (EOG) features provided by the SEED-VIG dataset. The feature pool employed in the study was enriched through feature engineering methods such as band ratios, temporal delta components, hemispheric asymmetry, and moving standard deviation. Moreover, by eliminating the need for raw signal processing, the focus was directed toward the temporal variation patterns of meaningful features. Experimental results demonstrated that predictions based solely on EEG data remained limited, whereas the inclusion of EOG data significantly improved performance.These findings reveal that EEG and EOG signals provide complementary information in the prediction of driver fatigue. The proposed multimodal approach exhibited greater robustness against inter-subject variability compared to unimodal systems and offered a more generalizable monitoring framework.
A multidimensional benchmarking framework for large language models in oncologic decision making
(Nature Research, 2026) Halıcı, Mehmet; Saltürk, Serkan; Sayın, İrem; Ertan, Burak; Balcı, İbrahim Cem; Çepni, Kimia; Kapağan, Tanju; Yıldırım, Cumhur; Erdem, Gökmen Umut; Kızıltan, Huriye Şenay; Koçak, Muhammed Tayyip; Üvet, Hüseyin
Large language models (LLMs) are increasingly explored as clinical decision support tools in oncology; however, reliance on isolated metrics has limited the development of multi-dimensional evaluation frameworks. This comparative observational study utilized five stepwise, clinically realistic non-small cell lung cancer scenarios reflecting real-world diagnostic, therapeutic, and follow-up decision making. Open-ended clinical questions were answered by three LLMs (Gemini 2.5 Pro, GPT-5, and Claude Opus 4.1) via their official APIs and compared with evidence-based reference answers. Model outputs were evaluated using expert-rated clinical accuracy and explainability, alongside operational metrics including cost, response time, and generative efficiency. All dimensions were integrated into an expert-weighted Composite Performance Score (CPS). Across 30 clinical questions, significant inter-model differences were observed for all metrics (p < 0.001). GPT-5 achieved the highest accuracy, explainability, and generative efficiency, while Gemini 2.5 Pro demonstrated the lowest cost and Opus 4.1 the fastest response times. Integrated analysis yielded the highest CPS for GPT-5, followed by Gemini 2.5 Pro and Opus 4.1 (Kendall’s W = 0.87). A multi-dimensional evaluation framework integrating clinical quality and operational efficiency provides more actionable insights than single metric assessments, enabling pragmatic model selection for oncology practice. Nevertheless, the use of LLMs in this domain should remain clinician-supervised.
Topology optimization of snake-like robot limb using ansys
(Fytronix Elektronik Teknolojileri, 2023) Coşkun, Yusuf; Aydın, Muhammet; Koçak, Muhammed Tayyip
Recent natural disasters lead to the formation of large amounts of debris and consequently to the trapping of living creatures among the debris. A search and rescue robot inspired by the morphology of a snake has been designed to effectively reach the trapped creatures in these challenging conditions. Optimization of the body limb is of great importance for the robot to perform fast, agile, and durable in narrow and complex areas under debris. In this study, a topology optimization analysis of the robot's limb is performed using Ansys software. The optimization is focused on the body limb, which is the basis of the snake's mobility. The main objective of this study is to achieve significant optimization of the mass and volume parameters for the robot's body limb while maintaining structural robustness. Thus, the overall design of the robot is aimed to be more compact and efficient, while achieving more effective mobility in narrow and difficult areas under debris. As a result of detailed analysis, a significant decrease in stress and strain values was found. However, an increase in deformation values was observed. The fact that this increase occurs in deformation indicates that some parameters of the model deviate from the expected results. Furthermore, the investigation revealed that the mass and volume of the model decreased by half. This shows the potential of topology optimization in terms of material utilization and structural efficiency. However, the increase in deformation indicates that the current parameters of the model should be reviewed. Consequently, it has become imperative to revise the selected parameters to further improve the performance of the model and achieve the expected mechanical properties. This study successfully demonstrates how lighter and more durable structures can be achieved through topology optimization with efficient use of resources.
Comparative benchmarking of 2d lidar slam algorithms with ros 2 on raspberry pi 5
(IEEE, 2026) Şahin, Ulaş; Can, Göktürk; Altıok, Ezgi; Çavdar, İbrahim; Gözüaçık, Necip
Mobile robotics increasingly relies on SLAM for robust autonomous navigation. While many algorithms exist, systematic comparisons within the ROS 2 framework under real-world conditions remain limited. This study addresses this gap by benchmarking three widely used 2D LiDAR-based methods—GMapping, Hector SLAM, and Cartographer—on a wheeled mobile robot. Using both simulation and on-device experiments, we evaluate mapping accuracy, localization quality, and computational efficiency. Results show that Cartographer achieves the highest accuracy in structured environments, Hector SLAM demonstrates robustness without odometry, and GMapping performs reliably only in small-scale settings. These findings highlight trade-offs relevant to embedded deployment. The main contributions are: (i) a reproducible evaluation pipeline on ROS 2, (ii) quantitative analysis of accuracy versus resource usage on Raspberry Pi 5, and (iii) practical guidelines for algorithm selection in autonomous systems. This work advances the understanding of ROS 2-based SLAM and supports informed deployment in robotics applications.
Multilevel compartment threshold secret image sharing scheme
(Zhubanov University, 2025) Nabiyev, Vasif; Soleymanzadeh, Katira
Traditional secret sharing schemes assume that all participants within a group or compartment possess equal authority in reconstructing the secret. However, in many real-world applications, such as hierarchical organizational structures or secure multi-party collaborations, this assumption does not hold. To address this limitation, we propose a novel Multilevel Compartment Threshold Secret Image Sharing (MCT-SIS) scheme that introduces hierarchical privileges within each compartment. Our scheme is based on a combination of Tassa’s hierarchical access structure and Ghodosi’s compartment model, and utilizes Birkhoff interpolation and polynomial-based techniques to achieve robust and flexible secret image sharing. Participants are grouped into disjoint compartments, each with multiple levels of access, and the secret image is shared such that it can only be reconstructed when both compartmental and hierarchical threshold conditions are satisfied. The scheme ensures perfect secrecy, lossless reconstruction, and reduced storage overhead. Experimental results validate its feasibility and demonstrate its applicability to environments requiring fine-grained access control, such as collaborative data vaults, medical imaging systems, and secure multi agency operations.
A comparative study of deep learning models for automated liver and tumor segmentation in 2d contrast-enhanced MRI images
(IEEE, 2025) Tokatlı, Nazlı; Bilmez, Yakuphan; Bayram, Mücahit; Bayır, Beyzanur; Özalkan, Helin; Tekin, Zeynep; Örmeci, Necati; Altun, Halis
This paper presents a comprehensive investigation into deep learning techniques for the automated segmentation of the liver and tumors from 2D abdominal contrast-enhanced Magnetic Resonance Imaging (MRI) slices. Addressing a significant challenge in medical image analysis, our study leverages the public ATLAS dataset [1], using a selection of 60 3D abdominal MRI scans, from which we extracted approximately 3,750 2D slices for model training and evaluation. The core objective was the precise identification and delineation of both the liver organ and any intrahepatic lesions. A comparative analysis was conducted on three U-Net-based architectures: the standard Attention U-Net model incorporating EfficientNet-b3 and CBAM but without Focal Loss, the Attention U-Net model with integrated Focal Loss, and the ResNet34-Based U-Net model. To optimize performance, we explored the efficacy of different loss functions, namely DiceLoss and a hybrid DiceLoss with Focalcoss. Our findings are promising: Among the evaluated models, the ResNet34-Based U-Net demonstrated the highest performance with a Dice score of 91.36% and an IoU score of 89.52%. It was followed by the Attention U-Net with Focal Loss, which achieved 86.41% Dice and 81.61% IoU scores, and the standard Attention U-Net, which obtained 85.93% Dice and 81.19% IoU scores. These results underscore the significant potential of our 2D-based methodology to enhance the precision and efficiency of liver and tumor detection from abdominal scans, offering a valuable tool to support clinicians in early diagnosis and to alleviate their workload.
An AI-powered mobile application for aroid identification and interactive learning: Enhancing pharmacognosy education through deep learning and NLP
(IEEE, 2025) Tokatlı, Nazlı; Bilmez, Yakuphan; Kılıç, Yusuf; Alpınar, Abdülkerim
Aroid plants (Araceae family), recognized for their distinct inflorescence, possess significant botanical, pharmaceutical, and practical importance due to their content of both beneficial compounds and toxins such as calcium oxalate crystals. Accurate identification of these species is particularly crucial in pharmacy education; however, morphological similarities among Aroid species often lead to confusion among students. This paper presents a deep learning-based mobile application designed to support both plant identification and interactive learning. The solution leverages EfficientNet and Convolutional Neural Network (CNN) architectures, achieving up to 96 % accuracy in classifying Aroid species. The visual classification model, trained on a comprehensive dataset, is deployed via a RESTful API and integrated within a Flutter-based mobile application. In addition, the app incorporates a Natural Language Processing (NLP)-powered chatbot to address user inquiries regarding plant characteristics and care. While technical evaluations demonstrate robust model performance, a comprehensive user evaluation aimed at assessing the system's educational value, usability, and chatbot interaction is planned as future work. This study underscores the potential of AI-driven mobile solutions in advancing pharmacognosy education, with future developments aimed at expanding the app's botanical scope and enhancing user engagement based on forthcoming survey results.
AI-powered digital assistant for chronic disease and elderly care management
(IEEE, 2025) Bayram, Hatice Merve; Gürcan, Zehra; Ayrıç, Esmanur; Gözüaçık, Necip
The increasing prevalence of chronic diseases and the aging population pose significant challenges to healthcare systems worldwide, necessitating innovative technological solutions to alleviate the burden. Previous studies have explored telemedicine and the integration of AI in healthcare, yet there remains a gap in comprehensive systems that integrate health monitoring, advisory services, and emergency alerts. This research aims to address this gap by developing an AI-powered digital assistant designed to enhance chronic disease management and elderly care. Utilizing a user-centered design approach, the study employs mobile health applications, AI-driven decision support systems, and a hybrid health tracking model that combines automatic and manual data entry. The system architecture includes a mobile application developed with Flutter, backend services using ASP.NET Core, and AI functionalities powered by Microsoft Azure's OpenAI models. Key findings demonstrate the system's effectiveness in improving user engagement in health management, providing timely alerts, and offering personalized health insights, thereby challenging existing assumptions about the limitations of digital health platforms. The study contributes to the field by offering a scalable, user-friendly solution that empowers individuals in managing their health, with practical implications for reducing healthcare dependency and enhancing patient autonomy. Future work will focus on expanding the system's capabilities and conducting real-world user testing to further refine its accessibility and usability.
AI-assisted medical image analysis and healthcare system integration
(IEEE, 2025) Akarçeşme, Furkan; Özcan, Halenur; Horata, Şerife Zülal; Edar, Yiğit; Gözüaçık, Necip
The integration of artificial intelligence (AI) in medical image analysis and healthcare systems is transforming modern medicine by enhancing diagnostic accuracy and reducing the burden on healthcare providers. Previous studies have demonstrated the potential of AI in medical imaging, yet there remains a need for comprehensive systems that seamlessly integrate Al-driven analysis with user-friendly healthcare platforms. This research aims to address this gap by developing a web-based health management system that facilitates online appointment scheduling and AI-assisted analysis of medical images and laboratory results. Utilizing a hybrid architecture, the system employs .NET for orchestration, Python microservices for data processing, and OpenAI's GPT for natural language interaction. The study involved testing the system with simulated data, achieving over 92% accuracy in radiological image analysis, and receiving positive feedback from users regarding its interface and functionality. These findings suggest that the system not only enhances diagnostic processes but also improves patient engagement and decision-making. The research contributes to the field by offering a novel, integrated platform that bridges the gap between technical data and patient understanding, with implications for future developments in digital health solutions. The system's design and successful implementation highlight its potential for real-world application, paving the way for further validation with clinical data and expansion into mobile platforms.
Braindetective: An advanced deep learning application for early detection, segmentation and classification of brain tumours using MRI images
(Springer Nature Link, 2025) Tokatlı, Nazlı; Bayram, Mücahit; Ogur, Hatice; Kılıç, Yusuf; Han, Vesile; Batur, Kutay Can; Altun, Halis
This study aims to create deep learning models for the early identification and classification of brain tumours. Models like U-Net, DAU-Net, DAU-Net 3D, and SGANet have been used to evaluate brain MRI images accurately. Magnetic resonance imaging (MRI) is the most commonly used method in brain tumour diag nosis, but it is a complicated procedure due to the brain’s complex structure. This study looked into the ability of deep learning architectures to increase the accuracy of brain tumour diagnosis. We used the BraTS 2020 dataset to segment and classify brain tumours. The U-Net model designed for the project achieved an accuracy rate of 97% with a loss of 47%, DAU-Net reached 90% accuracy with a loss of 33%, DAU-Net 3D achieved 99% accuracy with a loss of 35%, and SGANet achieved 99% accuracy with a loss of 20%, all demonstrating effective outcomes. These find ings aim to improve patient care quality by speeding up medical diagnosis processes using computer-aided technology. Doctors can detect 3D tumours from MRI pictures using software developed as part of the research. The work packages correctly han dled project management throughout the study’s data collection, model creation, and evaluation stages. Regarding brain tumour segmentation, 3D U-Net architecture with multi-head attention mechanisms provides doctors with the best tools for planning surgery and giving each patient the best treatment options. The user-friendly Turkish interface enables simple MRI picture uploads and quick, understandable findings.
A hybrid approach to credit risk assessment using bill payment habits data and explainable artificial intelligence
(MDPI Publishing, 2025) Bulut, Cem; Arslan, Emel
Credit risk is one of the most important issues in the rapidly growing and devel oping finance sector. This study utilized a dataset containing real information about the bill payments of individuals who made transactions with a payment institution operating in Turkey. First, the transactions in the dataset were analyzed based on the bill type and the individual and features reflecting the payment habits were extracted. For the target class, real credit scores generated by the Credit Registry Office for the individuals whose payment habits were extracted were used. The dataset is a multi-class, unbalanced, and alternative dataset. Therefore, the dataset was prepared for the analysis by using data cleaning, feature selection, and sampling techniques. Then, the dataset was classified using various classification and evaluation methods. The best results were obtained with a model consisting of ANOVA F-Test, SMOTE, and Extra Tree algorithms. With this model, 80.49% accuracy, 79.89% precision, and 97.04% UAC rate were obtained. These results are quite efficient for an alternative dataset with 10 classes. This model was transformed into an explainable and interpretable form using LIME and SHAP, which are XAI techniques. This study presents a new hybrid model for credit risk assessment based on a multi-class and imbalanced alternative dataset and machine learning.
AI-focused peer assessment in internship programs: Effects on virtual learning competencies and attitudes towards AI
(International Society for Technology, Education, and Science, 2025) Kayhan, Osman; Tokatlı, Nazlı; Altun, Halis; Korkmaz, Özgen
This research aims to investigate the impact of peer-evaluated internship activities centered on artificial intelligence on engineering students, specifically focusing on their virtual learning competencies, attitudes towards artificial intelligence, and ascertain student perspectives. For this purpose, mixed-methods research has been used. In this study, which used a sequential explanatory design, a preliminary experimental design was employed in the quantitative part and basic qualitative research methods were used in the qualitative part. The study group consists of 34 engineering students. Data were collected using the Project-Based Virtual Learning Competencies Scale, the General Attitude Towards Artificial Intelligence Scale, and semi-structured interview forms. The research demonstrated that internship activities centered on peer evaluation related to artificial intelligence in virtual learning environments considerably improved the virtual learning competencies of engineering students. However, these activities did not significantly influence their attitudes towards artificial intelligence, whether positive or negative. The tasks during the internship program are thought to have improved students' collaboration, problem-solving, communication, creative thinking, technical knowledge, project writing skills, and research skills. Furthermore, it has been determined that students have acquired experience and enhanced their skills in process management, leadership, and tolerance.
DeepMatch: A BERT-powered talent matchmaking approach
(Springer Nature Link, 2025) Gözüaçık, Necip; Topaloğlu, Atakan; Evren, Ayse Mine; Karakuş, Serkan; Akram Bennour; Ahmed Bouridane; Somaya Almaadeed; Bassem Bouaziz; Eran Edirisinghe
Consultancy companies aim to match their employees to customer projects based on their employee’s talents. Traditional matchmaking methodologies are founded on manual processes that rely on rules of thumb or algorithms that are based on handcrafted heuristics, which cause the matchings to be not only sub-optimal, but also time-consuming, subjective, and prone to human errors. In this paper, we propose a novel consultancy matching algorithm that utilizes BERT to semantically find the most optimal consultant-project matchings for a given set of consultants and projects, pairing relevant project specifications with consultant specifications using the JVSAP algorithm. In doing so, our proposed talent matchmaking system may be utilized to improve the accuracy and efficiency of consultancy matching, thereby facilitating more effective consultancy engagements. Our findings suggest that the pairings demonstrate a discernible alignment with human intuition, as evidenced by the consistent correlation between consultants possessing domain-specific expertise and projects characterized by corresponding thematic descriptions. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.
MelanoTech: Development of a mobile application infrastructure for melanoma cancer diagnosis based on artificial intelligence technologies
(IEEE, 2024) Tokatlı, Nazlı; Bilmez, Yakuphan; Göztepeli, Gürkan; Güler, Muhammed; Karan, Furkan; Altun, Halis
This preliminary work introduces MelanoTech, a mHealth application designed and implemented to offer a user-friendly and intuitive interface for the early diagnosis of melanoma, a kind of skin cancer with significant fatality rates [1]. The application demonstrates promising performance in segmentation and classification tasks by utilizing deep learning models with Generative Adversarial Networks (GANs) for data augmentation. MelanoTech achieves a comprehensive accuracy rate of 92%, with a segmentation model accuracy rate of 93% and a lesion detection accuracy rate of 90%. Finally, incorporating data augmentation approaches based on GANs resulted in a 5% enhancement in the model’s performance. These findings highlight the capacity of MelanoTech as a dependable tool for improving the early diagnosis of melanoma and decreasing the workload of physicians in Turkish public hospitals.
Log-Harmonic mappings associated with the sine function
(2025) Kumar, Sushil; Çetinkaya, Asena; Özkan Uçar, Hatice Esra
In this paper, we define new subclasses STlh(s) and CSTlh(s) of sine starlike log-harmonic mappings and sine close-to-starlike log-harmonic mappings, respectively, defined in the open unit disc D. We investigate representation theorem and integral representation theorem for functions in the class STlh(s). Further, we determine radius of starlikeness for functions in the classes STlh(s) and CSTlh(s).
A study on harmonic functions
(Bulgarian Academy of Sciences, 2024) Uçar, Mehmet Fatih; Özkan Uçar, Hatice Esra
The class of functions that have bounded boundary rotation and bounded radius rotation are the generalization of the convex and the starlike functions, respectively. The concept of such functions was introduced by L¨owner [1]. But he did not use the present terminology. It was Paatero [2,3] who systematically developed their properties and made an exhaustive study of the class of functions that have bounded boundary rotation. We will examine in this paper, the subclass of SH that is related to the class of functions that have bounded radius rotation.
Artificial intelligence-based fair allocation in NOMA technique: A review
(Bentham Science Publishers, 2024) Kırtay, Seda; Yıldız, Kazım; Böcekçi, Veysel Gökhan
Non-Orthogonal Multiple Access (NOMA) is an innovation that has great potential in wireless communication. It permits multiple users to efficiently allot a frequency band by adjusting their power allocations. Nevertheless, attaining fair power allocation in NOMA structures presents complex challenges that require specific models, extensive training data, and addressing issues of generalization. This review aims to explore the applications of Artificial Intelligence (AI) and Deep Learning (DL) methods to tackle the challenges associated with fair power allocation in NOMA systems. The focus is on developing strong AI-DL models and creative optimization methods specifically designed for dynamic environments to improve transparency and interpretability. This study explores a wide range of techniques, including Reinforcement Learning, Convolutional Neural Networks (CNN) for power allocation, Generative Adversarial Networks, Deep Reinforcement Learning, and Transfer Learning. The goal is to enhance various aspects, such as power allocation, user coupling, scheduling strategies, interference cancellation, user mobility, security, and deeplearning- based NOMA. Despite the difficulties, impartial power allocation algorithms based on AI and DL show promise in improving user performance and promoting fair power distribution in NOMA systems. This study emphasizes the significance of continuous research efforts to overcome current obstacles, enhance efficiency, and strengthen the dependability of wireless communication systems. This highlights the significance of NOMA as an advanced innovation for upcoming wireless generations that go beyond 5G. Future areas of study involve investigating federated learning and novel techniques for gathering data and utilizing interpretable AI-DL models to address existing constraints. Overall, this review highlights the potential of AI and DL techniques in achieving fair power distribution in NOMA systems. However, further investigation is crucial to addressing obstacles and fully exploring the capabilities of NOMA technology.
Transfer learning in severity classification in Alzheimer's : A benchmark comparative study on deep neural networks
(Altınbas University, 2024) Kırtay, Seda; Koçak, Muhammed Tayyip
Alzheimer's disease has become a condition of the brain that progresses over time and impacts a significant number of individuals worldwide. Early diagnosis, timely intervention and management of this disease process are very important in Alzheimer's disease. With regard to this study, we propose a transfer learning based early detection approach for Alzheimer's disease using Moderate Demented, Mild Demented, No Demented and Very Mild Demented classification sets. The proposed approach utilizes transfer learning based on the use of a deep neural network model that has been trained to extract features from brain imaging data. To evaluate the performance in transfer learning, a dataset of 6,400 images from brain MRI scans is augmented using data augmentation techniques and used in various convolutional neural network models the like VGG-19, Resnet-50, DenseNet-121, Inception-V3, VGG-16. The results are planned to show that these models achieve high sensitivity, specificity and high accuracy in detecting early signs of Alzheimer's disease. The study also emphasizes these advantages of using transfer methods of learning for early Alzheimer's detection by comparing it with various other deep learning models. The findings of this research suggest that transfer learning-based approaches can aid in the early detection of Alzheimer's disease., which affects millions of people, and offer a practical solution to classify cognitive impairment. With the proposed approach, it is shown that by helping clinicians to detect individuals at risk of Alzheimer's at an early stage, it will be possible to provide timely intervention and, in fact, better patient care. In terms of more effective applicability in clinical applications, the proposed approach can be applied to different and larger datasets and populations to make improvements and provide convenience to clinicians and patients. The best success rate of the models we used is achieved on the VGG19, RESNET50 KNN model with 99 percent.
Nuclei instance segmentation in colon histology images with YOLOv7
(Springer Nature, 2024) Yıldız, Serdar; Memiş, Abbas; Varlı, Songül
In histology image analysis, instance-based nuclei segmentation is one of the challenging tasks within the segmentation-guided studies since it is quite troublesome to detect each distinct nuclei instance of each nuclei type in images in contrast to the semantic segmentation in which all the image pixels of a nuclei type are labelled with the same mask ID although the segmented region may comprise of multiple instances. In this paper, an instance-based medical image segmentation task is addressed, and in this context, instances of multiple types of nuclei in colon histology images are aimed to be delineated distinctly. For the instance-based segmentation of the nuclei in colon histology images, the YOLOv7 algorithm and its built-in instance segmentation module are utilized. In the experimental studies performed on Colon Nuclei Identification and Counting (CoNIC) Challenge 2022 colon histology image dataset by using a 5-fold cross-validation performance evaluation strategy, nuclei instances belonging to 6 classes as the neutrophil, epithelial, lymphocyte, plasma, eosinophil and connective were segmented. To calculate the overall system accuracy, the quantification metrics of mean average precision (mAP) and mean panoptic quality (mPQ) were measured. In performance evaluations, quite promising accuracy values were obtained. The mAP values of 0.2885 and 0.2903, and mPQ values of 0.1659 and 0.1704 were observed by using the YOLOv7 algorithm. To the best of our knowledge, this is the first nuclei instance segmentation study with YOLOv7.

Güncel Gönderiler