Doktora Tezleri

Permanent URI for this collectionhttps://hdl.handle.net/20.500.12573/5800

Browse

Now showing 1 - 20 of 22

Anormallik Tespiti için Veri Madenciliği
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2020) Kaçmaz, Rukiye Nur; Yılmaz, Bülent
Gastroentereloji uzmanları için kolon anormalliklerinin tespit edilmesi en zor görevlerden birisidir. Kolonoskopi herhangi bir anormalliği izlemek için kolondan video veya görüntüler kaydetmenin en yaygın yöntemidir. Bununla birlikte işlem sırasında elde edilen görüntü veya videolar, kolonoskopi probunun ya da kapsülün hızlı hareketinden kaynaklanan hareket gürültüsü, kapsülde ve probda ışık kaynağından kaynaklanan yansıma gürültüsü (YG), yetersiz veya aşırı aydınlatmadan kaynaklanan uygun olmayan kontrast gürültüsü, mide öz suyu, baloncuklar veya kalıntılar içermektedir. Bu tarz gürültüler içeren görüntülere bilgi taşımayan çerçeveler adı verilmektedir. Hastalık tespiti işlemi ise bilgi içeren olarak adlandırılan temiz görüntüler ile yürütülmektedir. İlk çalışmada tekstür tabanlı otomatik polip tespitinde YG'nin etkisini ve YG'yi ortadan kaldırmak için kullanılan görüntü enterpolasyonunun kullanımı araştırıldı. Bu amaçla, çeşitli boyutlarda sonradan YG eklenen ve interpolasyon uygulanan görüntülerden ve YG içermeyen görüntülerden çeşitli tekstür özellikleri elde edildi. Polipleri kolon arka planından ayırt etmek için, uygulanan en yakın komşular, bilineer ve bikübik interpolasyon yöntemlerinin, tekstür özellikleri ve sınıflandırma performansı açısından herhangi bir farklılığa neden olup olmadığı test edildi. İkinci çalışmada temel amaç, bilgi taşımayan çerçeveleri tespit etmede geleneksel makine öğrenmesi ve transfer öğrenme yaklaşımlarının performanslarının karşılaştırılmasıydı. Makine öğrenmesi bölümünde, gri seviye eş oluşum matrisi, gri seviye koşu uzunluğu matrisi, komşuluk gri ton farkı matrisi, odak ölçüm operatörleri ve basıklık, standart sapma ve çarpıklık olarak üç adet birinci derece istatistik kullanıldı. Sınıflandırma aşamasında rastgele orman, destek vektör makineleri ve karar ağacı yaklaşımları kullanılmıştır. Transfer öğrenme bölümünde derin sinir ağları olarak AlexNet, SqueezeNet, GoogleNet, ShuffleNet, ResNet-18, ResNet-50, NasNetMobile ve MobileNet tercih edildi. Son çalışma, bilgi taşıyan çerçevelerde Crohn's, ülseratif kolit, kanser ve polip gibi kolon anormalliklerinin saptanmasını içermiştir. Bu çalışmanın amacı, öncelikle sağlıklı çerçeveleri hastalıklılardan ayırmak ve hem geleneksel makine öğrenmesi hem de transfer öğrenme yaklaşımlarını kullanarak hastalık türlerini belirlemekti. İkinci çalışmada kullanılanlarla aynı tekstür özellikleri, sınıflandırma yaklaşımları ve transfer öğrenme yöntemleri kullanılmıştır.
Biyoçipler için Mikro Biyomalzemelerin ve Hücrelerin Görüntü İşleme Yöntemleri ile Otomatik Olarak Sayılması ve Analizi
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2023) Çelebi, Fatma; İçöz, Kutay
Quantification of tumor cells is essential for early cancer detection and progression tracking. Multiple techniques have been devised to detect tumor cells. In addition to conventional laboratory instruments, several biochip-based techniques have been devised for this purpose. Our biochip design incorporates micron-sized immunomagnetic beads and micropad arrays, necessitating automated detection and quantification not only of cells but also of the micropads and immunomagnetic beads. The primary function of the biochip is to simultaneously acquire target cells with distinct antigens. As a readout technique for the biochip, this study devised a digital image processing-based method for quantifying leukemia cells, immunomagnetic beads, and micropads. Images were acquired on the chip using bright-field microscopy with image objectives of 20X and 40X. Conventional image processing methods, machine learning methods, and deep learning methods were used to analyze the images. To quantify targets in the images captured by a bright-field microscope, color- and size-based object recognition and machine learning-based methods were first implemented. Secondly, color- and size-based object detection and object segmentation methods were implemented to detect structures in bright-field optical microscope images acquired from the biochip. Third, segmentation of the minimal residual disease (MRD) using deep learning. Implemented biochip images comprised of leukemic cells, immunomagnetic beads, and micropads. Moreover, mesenchymal stem cells (MSCs) are stem cells with the capacity for multilineage differentiation and self-renewal. Estimating the proportion of senescent cells is therefore essential for clinical applications of MSCs. In this study, a self-supervised learning (SSL)-based method for segmenting and quantifying the density of cellular senescence was implemented, which can perform well despite the small size of the labeled dataset.
Blokzincir Tabanlı Eşten-Eşe Enerji Ticareti Uygulamaları
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2023) Seven, Serkan; Alkan, Gülay Yalçın
This thesis explores the potential of innovative peer-to-peer (P2P) energy trading schemes for virtual power plants (VPPs) using blockchain technologies, smart contracts, and decentralized finance (DeFi) instruments. Traditional centralized approaches have limitations in terms of transparency and security, which can hinder the successful implementation and operation of VPPs and P2P energy trading systems. The dissertation begins by reviewing the current state of energy sources within the global energy landscape. Understanding the existing landscape provides valuable insights into the potential benefits and challenges of implementing P2P energy trading within VPPs. The focus of the dissertation is to develop and analyze innovative P2P energy trading schemes for VPPs that integrate blockchain technologies and facilities to enhance transparency, security, and automation of energy transactions. Furthermore, DeFi instruments, specifically decentralized exchange (DEX), are used as a novel approach instead of auction methods to determine P2P energy buying and selling prices. Along with blockchain technologies, optimization is used to maximize the economic benefits of peers. The sequential decision problem of the trading schemes is solved with mixed integer linear programming (MILP). In addition, machine/deep learning models are utilized to overcome the drawbacks of conventional mathematical programming like MILP. These models can accelerate the decision-making processes by learning from the optimization results obtained. Overall, frameworks for the successful integration of P2P energy trading within and among VPPs are developed to validate the effectiveness and feasibility of the proposed P2P energy trading schemes through case studies and simulations using realistic data sets and blockchain platforms.
Derin Öğrenme Tabanlı Kompozit Malzemelerin Ultrasonik Tomografi Görüntülerinden Kusurların Tespiti ve Sınıflandırılması
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2024) Gülşen, Abdulkadir; Güngör, Burcu; Kolukısa, Burak
This thesis introduces novel methodologies for enhancing defect classification and characterization in advanced composite materials by leveraging state-of-the-art machine learning (ML), deep learning (DL), and federated learning (FL) techniques within ultrasonic and acoustic emission (AE) inspection environments. First, a new ultrasonic dataset (UNDT), comprising 1,150 images from 60 distinct composite materials, is introduced. Applying transfer learning methods to both the UNDT and a publicly available dataset demonstrates the efficacy of advanced neural architectures—such as DenseNet121 and VGG19—achieving accuracy rates up to 98.8% and 98.6%, respectively. Next, the scope is extended to AE-based health monitoring by introducing an ensemble feature selection methodology to identify features strongly correlated with damage modes. By selecting amplitude and peak frequency for labeling and subsequently applying unsupervised clustering, the analysis confirms that both traditional AE features (e.g., counts and energy) and less commonly employed features (e.g., partial powers) correlate with distinct defect types. Finally, a novel FL framework is introduced to address the scarcity of publicly available, real-world ultrasonic datasets. This decentralized approach preserves data privacy while maintaining performance levels comparable to centralized methods, ensuring scalability and confidentiality in diverse data environments. Overall, these contributions significantly advance the field of NDT, offering robust defect classification and characterization. In doing so, the findings not only improve the accuracy and reliability of material integrity assessments but also lay a durable foundation for more secure, collaborative, and efficient NDT systems.
Derin Öğrenme Yaklaşımlarıyla Küçük Hücreli Dışı Akciğer Kanserinde Tümör Karakterizasyonu
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2021) Bıçakcı, Mustafa; Yılmaz, Bülent
Küçük Hücreli Dışı Akciğer Kanseri (KHDAK) akciğer kanserlerinin büyük çoğunluğunu oluşturur ve adenokarsinom (ADC) ve skuamöz hücreli karsinom (SqCC) olmak üzere iki önemli alt tipi vardır. Genel olarak, bu iki alt tip mikroskobik olarak belirlenen morfolojik kriterler dikkate alınarak birbirinden ayrılır. Ancak, kötü morfoloji bunu oldukça zorlaştırır. Alt tipe özel tedavi yöntemleri için bu tür çalışmalar önemlidir. Bu tezde, pozitron emisyon tomografi (PET) görüntüleri kullanılarak KHDAK'nin alt tiplerinin sınıflandırılması üzerinde derin öğrenme (DÖ) yöntemleri incelenmiştir. İlk çalışmada, DÖ yöntemlerinin temelini oluşturan yapay sinir ağları (YSA) kullanılarak %73 doğru sınıflandırma başarısı elde edilmiştir. İkinci çalışmada, PET görüntülerinden alınan bölütlenmiş tümör kesitleri kullanılarak birkaç DÖ modeli incelenmiştir. Sonuçta, %95 F skoru ile VGG16 ve VGG19 en başarılı modeller olmuştur. Bu çalışmanın sonunda kesit bazlı çalışmalar bırakılarak hasta bazlı çalışmalara geçilmiştir. Üçüncü çalışmada, hasta bazlı dilimlerin birleştirilmesiyle oluşturulan üç boyutlu (3B) verilerin kullanımı yeterli başarıyı sağlamamıştır. Dördüncü çalışmada, PET görüntülerinin doğrudan kullanıldığı, tümör kısımlarının kırpılarak kullanıldığı ve bölütlenmiş tümör parçalarının kullanıldığı üç farklı deney yapılmıştır. Bu çalışma, peritümoral alanların sınıflandırmada olumlu etkisini ortaya koymuş ve VGG19 %74 F skoru değerine ulaşmıştır. Beşinci çalışmada, transfer öğrenme ve hassas ayar çalışmaları başarısızdı. CNN ve ResNet tabanlı sığ ağları içeren son çalışma %71 F skoru ile umut verici olmuştur.
EEG Sinyallerinden Disfaji Hastalığının Karakteristiklerinin Belirlenmesi ve Analizi
(2025) Aslan, Sevgi Gökçe; Yılmaz, Bülent
Disfaji, genellikle nörolojik hastalıklarla ilişkilendirilen ve özellikle yaşlı bireylerde yaşam kalitesini olumsuz yönde etkileyen bir yutma bozukluğudur. Bu çalışma, EEG verileri kullanılarak yutma ve yutmayı hayal etme süreçlerinin nörofizyolojik analizini yapmayı ve bu verilerin disfaji rehabilitasyonunda nasıl kullanılabileceğini araştırmaktadır. Otuz adet sağ elini kullanan birey üzerinde gerçekleştirilen deneylerde, doğal yutma, indüklenmiş tükürük yutma, indüklenmiş su yutma ve indüklenmiş dil dışarı çıkarma gibi farklı deneysel paradigmalar kullanılmıştır. Verilerin ön işlenmesinde Bağımsız Bileşen Analizi (ICA), Empirik Mod Ayrıştırma (EMD), bant geçiren filtreleme ve Ortak Uzamsal Desen (CSP) analizi gibi teknikler uygulanmıştır. Bu ön işleme yöntemleri, EEG verilerindeki gürültüyü azaltarak daha doğru bir analiz sağlamak amacıyla kullanılmıştır. Geleneksel makine öğrenmesi teknikleri ve derin öğrenme yöntemleriyle yapılan sınıflandırma görevlerinde, dinlenme ve hayal etme evreleri arasındaki farklar belirgin bir şekilde ayrılmıştır. Random Forest, AdaBoost ve Bagging gibi topluluk tabanlı algoritmaların yanı sıra, derin öğrenme yöntemlerinden Konvolüsyonel Sinir Ağları (CNN) da uygulanmıştır. Ayrıca, çok ölçekli mekânsal dikkat ağı (MS-SAN) modeli, özellikle delta ve teta frekans bantlarında hareketi hayal etme ile dinlenme durumları arasındaki nörofizyolojik farkları yüksek doğrulukla ayırt etmiştir. Sonuçlar, hareketi hayal etme ve dinlenme evrelerinin EEG verileriyle tespit edilmesinin disfaji tedavisinde ve motor rehabilitasyon uygulamalarında büyük bir potansiyel taşıdığını göstermektedir. Bu çalışma, EEG tabanlı beyin-bilgisayar arayüzleri (BBA) teknolojilerinin, makine öğrenimi ve derin öğrenme yöntemlerinin disfaji rehabilitasyonundaki potansiyelini vurgulamakta ve bu alandaki araştırmaların klinik uygulamalar açısından önemini ortaya koymaktadır. Anahtar kelimeler: Elektroensefalografi, Makine Öğrenmesi, Derin Öğrenme, BBA, Yutkunma
Endüstriyel Ortamlarda Enerji Hasatlayan Çoğul Ortam Kablosuz Algılayıcı Ağları
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2020) Tekin, Nazlı; Güngör, Vehbi Çağrı
Sert kanal koşullarına sahip olan Endüstriyel Kablosuz Algılayıcı Ağ'larda (EKAA), enerji verimli ve güvenilir kablosuz iletişim sağlamak büyük önem taşımaktadır. Ağ güvenirliğini sağlarken aynı zamanda ağın ömrünü uzatmak da zor bir problemdir. Bu çalışmanın amacı, EKAA'ların ömrünün eniyilenmesidir. Bunu yaparken, endüstriyel ortamlar için uygun olan iç mekan güneş, termal ve titreşime dayalı Enerji Hasatlama (EH) yöntemleri tanımlanmış ve bunların ağ ömrüne katkıları araştırılmıştır. Uygulama güvenilirliğini ve EH yöntemlerini birlikte değerlendirerek, ağ ömrünü eniyilemek için yeni bir Karma Tamsayılı Programlama (KTP) modeli formüle edilmiştir. Ayrıca, Kablosuz Çoğul Ortam Algılayıcı Ağ'larında (KÇOAA) iletişim, büyük veri boyutu nedeniyle fazladan enerji tüketimine sebep olur. Bu nedenle, büyük veri boyutunu iletimden önce azaltmak önemli hale gelir.Bu amaçla, iletişim ve enerji dağıtım hesaplamalarını dikkate alırken, sıkıştırıcı algılama ve görüntü sıkıştırma gibi veri boyutu küçültme yöntemlerinin endüstriyel ağ ömrü üzerindeki etkisi değerlendirilir. Öte yandan, özellikle çok sayıda algılayıcılar bulunduran ağlar için KTP modelini uygun bir zamanda çözmek bir hayli zordur. KTP'nin zaman karmaşıklığı sorununun üstesinden gelmek için sezgisel tabanlı yöntemler geliştirilmiştir.
FDG-PET Görüntülerindeki Tümörlerin Makine ve Derin Öğrenme Tabanlı Analizi
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2022) Ayyıldız, Oğuzhan; Yılmaz, Bülent
Analysis of a tumor is essential in treatment planning and evaluation of treatment response. Positron Emission Tomography (PET) is a vital imaging device for clinical oncology in understanding the metabolic structure of the tumor. In this thesis, three separate studies investigating the application of machine, deep learning and statistical approaches on FDG-PET images from patients with non-small cell lung cancer (NSCLC) and pancreatic cancer. The first study aimed at performing a survey on subtype classification of NSCLC by using different texture features, feature selection methods and classifiers. Images from 92 patients and several clinical and metabolic features for each case were used in this study along with histopathological validation for the tumor subtype labeling. Stacking classifier resulted in 76% accuracy. The aim of our second study was to adapt an atrous (dilated) convolution-based tumor segmentation approach (DeepLabV3) on FDG-PET slices with maximum standard uptake value (SUVmax). MobileNet-v2 pretrained on ImageNet served as the backbone to DeepLabV3. The classification layer was interchanged with the Tversky loss layer which helped improve model's performance while the dataset was imbalanced. Images from 141 patients were employed and augmentation was performed in each training phase. Dice similarity index was obtained as 0.76 without preprocessing and 0.85 with preprocessing. The last study focused on determining the features to be used in the prognosis of pancreatic adenocarcinoma on FDG-PET images from 72 patients. Well-known texture, metabolic and physical features were extracted from tumor region that was determined with the help of random walk segmentation algorithm. On these features time-dependent ROC curve analysis was performed for 2-year overall survival (OS) prediction, and, in the univariable analyses, tumor size, energy, entropy, and strength were found to be significant predictors of OS. Keywords: PET/CT, NSCLC, Machine learning, Deep learning, Radiomics, Semantic segmentation
Genetik ve Enfeksiyon Hastalıklarının Tespiti için Makine Öğrenmesi Yöntemleri
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2024) Işık, Yunus Emre; Aydın, Zafer
Completion of the whole human genome in the 2003 has led to various advances in many fields, particularly in biology, genetics, health sciences, treatment, and pharmacology. In the following years, spread of faster and cheaper sequencing technologies has enabled us to extract and analyze genetic profiles of individuals digitally. Consequently, individual-specific forecasting and personalized treatment and precision medicine-, what once seemed like science fiction, have become more and more real. In both approaches, one of the crucial steps is identifying the presence of diseases using individual-specific genetic data. This thesis aims to comprehensively and comparatively evaluate the predictive performance of machine learning methods for Behçet's disease and respiratory infections. Additionally, feature selection methods were employed to identify the genetic factors (such as SNPs and genes) associated with disease presence for both diseases. Furthermore, the usability of selected features depending on biological pathway-driven active subnetworks listed in the literature was analyzed for the prediction of Behçet's disease. For the respiratory infection prediction problem, on the other hand, the prediction performance of features calculated by single-sample gene set enrichment analysis (ssGSEA) was evaluated using different machine learning methods. As the data types used in both experiments were different (genome-wide association studies data, gene expression profiles), the performance of machine learning approaches on different data types was also observed. It is hoped that the findings of both experiments will contribute to future machine learning based disease prediction studies.
Görüntü Tanıma Tabanlı Gökyüzü Kamerası Entegrasyonunu Kullanarak Sezgisel Vektörize Öğrenme Yöntemine Dayalı PV Tahmini
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2023) Yavuz, Levent; Önen, Ahmet
In order to ensure the continuity of demand and production balance, the use of renewable energy resources (RES) by countries will increase in the near future. Solar power generation is important for the integration of renewable energy into the power grid, but it can cause problems in power systems due to the uncertain and intermittent nature of solar power. Deep learning methods provide promising results in solar energy prediction, but the performance of these models depends on the initial weights assigned to the network. In this thesis, a novel weight initialization method, the Heuristic Vectorised Learning method, which takes into account certain characteristics of solar generation data has been proposed. This method aims to achieve better accuracy in solar forecasting by combining a statistical approach with a method based on deep learning. The method was compared with other commonly used methods such as Xavier, LeCun, He and Random, and it was seen that the proposed method performed better. Overall, the proposed weight initialization method provides significant benefits for solar forecasting applications in the context of renewable energy integration into the power grid. So, to reach higher accuracy, monitoring the sky is the best option for intra-day forecasts. Therefore, a hybrid model was created for photovoltaic generation prediction by using it together with environmental sensor data. The proposed method and panel shading model achieve higher accuracy values at the Abdullah Gül University campus in Kayseri. The proposed system provides a reliable PV energy forecast for the intraday energy markets.
Hastalık Tahmini ve Biyobelirteçlerin Tespiti için Makine Öğrenim Modellerinin Tasarımı ve Geliştirilmesi
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2024) Temiz, Mustafa; Güngör, Burcu; Yousef, Malik
In medical science, the prediction of diseases and the identification of biomarkers play an important role in the diagnosis and treatment of various health conditions. The recent proliferation of data mining techniques has accelerated the development of disease prediction systems. In particular, machine learning methods are an effective way to analyze medical data and identify patterns to predict the likelihood of the disease development. Machine learning methods also help to identify biomarkers. Recently, the increasing incidence and mortality rates of inflammatory bowel disease, colorectal cancer and type 2 diabetes have drawn researchers' attention to these research areas. The aim of this thesis is to reduce the number of features and improve the prediction performance of machine learning based on complex biological datasets with a large number of disease-related features, as well as to identify potential biomarkers. In this thesis, three different studies are presented. The first study predicts eleven different cancer subgroups using miRNA data and biological domain knowledge and identifies potential biomarkers for these diseases. The second study predicts three different diseases using metagenomic data and biological domain knowledge and identifies potential biomarkers. The third study uses metagenomic data related to colorectal cancer to conduct global and population-based comprehensive experiments with traditional feature selection methods to identify potential biomarkers. This thesis presents a promising avenue for early disease detection, facilitating expedited treatment protocols, improving human survival rates, and potentially alleviating economic burdens within these critical research domains.
Histopatoloji Görüntülerinden Bilgisayar Destekli Kanser Tespiti
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2023) Taşdemir, Sena Büşra Yengeç; Yılmaz, Bülent; Aydın, Zafer
Detecting colon adenomatous polyps early is crucial for reducing colon cancer risk. This thesis investigated various deep learning approaches for computer-aided diagnosis of colon polyps on histopathology images using deep learning. The thesis addressed key challenges in polyp classification, including differentiating adenomatous polyps from non-adenomatous tissues and multi-class classification of polyp types. Initially, a histopathology image dataset is collected and refined from Kayseri City Hospital. The first study used stain normalization algorithms and an ensemble framework for binary classification, achieving 95% accuracy on the custom dataset and 91.1% and 90% on UnitoPatho and EBHI datasets, respectively. The second study implemented a tailored version of the supervised contrastive learning model for multi-class classification, outperforming state-of-the-art deep learning models with accuracies of 87.1% on custom dataset and 70.3% on UnitoPatho dataset. The third study proposed a self-supervised contrastive learning approach for utilizing all data in cases of limited labeled images. This approach achieved better performance than transfer learning with ImageNet pre-trained models. In conclusion, this PhD thesis investigated deep learning approaches for computer-aided diagnosis of colon polyps on histopathology images, demonstrating high accuracy in binary and multi-class classification, outperforming state-of-the-art models. These findings contribute to improving colon polyp classification accuracy and efficiency, ultimately facilitating the early detection and prevention of colon cancer.
IPMSM’in HF Sinyal Enjeksiyonu ve Kayan Kipli Gözlemci Tabanlı Sensörsüz Kontrolü
(2024) Ateş, Ertuğrul; Tekgün, Burak; Barut, Murat
Bu çalışmada, gömülü mıknatıslı senkron makinede (GMSM) genişletilmiş elektromotor kuvveti (GEMK), rotor pozisyonu ve rotor hızının gerçek zamanlı kestirimi için faz kilitli döngü (FKD) ile birleştirilmiş yüksek frekans (YF) gerilim sinyali enjeksiyonuna dayalı bir kayma modlu gözlemci (KMG) tanıtılmaktadır. Bu yaklaşım, özellikle düşük hızlarda ve durma anında rotor pozisyonu ve hız kestirimlerinde zorlanan geleneksel KMG ve FKD tekniklerinin sınırlamalarını ele almak üzere tasarlanmıştır. GMSM kontrolünde bu durumlar, rotor pozisyonu tespiti için kritik olan zıt EMK sinyallerinin zayıflama veya belirsiz hale gelme eğiliminde olması nedeniyle önemli zorluklar ortaya çıkarır, bu da geleneksel yöntemler kullanıldığında yanlış kestirimlere yol açar. Bu sorunları çözmek için önerilen KMG, makineyi uyararak motorun gerçek hızına daha az bağımlı olan belirgin GEMK sinyalleri üreten YF gerilim enjeksiyonundan yararlanır. Bu yenilik, sıfır veya sıfıra yakın hızlarda dahi tutarlı ve gürültüye dayanıklı GEMK kestirimine olanak tanıyarak rotor pozisyonu ve hızının doğru şekilde kestirilmesi için bir temel oluşturur. FKD, bu GEMK kestirimlerini rafine ederek rotorun hız ve pozisyon bilgilerinin hassas bir şekilde elde edilmesini sağlar. GEMK sinyaliyle stabil bir faz kilidini koruyarak, FKD gürültüyü filtreler ve rotor pozisyonu ve hız ölçümlerinin doğruluğunu artırır. Bu temel üzerine, önerilen KMG-FKD kombinasyonunu kullanarak GMSM için sensörsüz hız kontrol sistemi geliştirdik ve uyguladık. Gerçek zamanlı sistem, düşük hız ve durma durumları dahil olmak üzere geniş bir çalışma aralığında test edilmiştir. 8 kutuplu, 0,4 kW'lık bir GMSM motorundan elde edilen deneysel sonuçlar, önerilen yöntemin geleneksel KMG ve FKD tekniklerine kıyasla üstün verimlilik ve sağlamlığını doğrulamaktadır.
Makine Öğrenmesi Yöntemleriyle Antimikrobiyal Peptit Aktivite Tahmini
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2023) Söylemez, Ümmü Gülsüm; Güngör, Burcu
Antimicrobial peptides (AMPs) are considered as promising alternatives to conventional antibiotics in order to overcome the growing problems of antibiotic resistance. Computational prediction approaches receive an increasing interest to identify and design the best candidate AMPs prior to the in vitro tests. In this thesis, using the multiple properties of the peptides we aimed to develop machine learning approaches that can predict the antimicrobial activities of the peptides. We have created two datasets for the peptides showing antimicrobial activity against Gram-negative and against Gram-positive bacteria separately. In our first study, ten different physico-chemical properties of the peptides were calculated, and used as features of the peptides. Following the data exploration and data preprocessing steps, a variety of classification models were build with 100-fold Monte Carlo Cross-Validation; and the performance of these models were evaluated. In the second study, we proposed a novel method called AMP-GSM. The method was tested for three datasets, and the prediction performance of AMP-GSM was comparatively evaluated with several feature selection methods and several classifiers. In the last study, using motif matching score with the models of activity against Gram-positive and Gram-negative bacteria, we created novel peptides and predicted the target selectivity of these peptides. The studies presented in this thesis advance the field of computational research by making it easier to predict the possible antimicrobial effects of peptides and to design AMPs in wet laboratory environments.
Medikal Termal Görüntülerin Otomatik Olarak İşlenmesi ve Sınıflandırılması
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2022) Özdil, Ahmet; Yılmaz, Bülent
The aim of this dissertation is to develop computer aided methods for processing and evaluating medical infrared thermal images. Throughout this study three problems were evaluated. The first problem was to automatically classify the body part and pose in the thermal images. In this study there were four classes; upper-lower body parts with back-front views. The first step included the segmentation of the background with Otsu's thresholding method applying histogram equalization. Next, DarkNet-19 architecture was used to extract features from images and these features were reduced using PCA and t-SNE methods. Finally reduced feature sets were used for classification. The second problem was to automatically classify liver steatosis from using thermal images. In this study, the classification problem was tested on an anatomical region of interest from abdominal images corresponding to the liver. Deep learning and texture analysis methods were employed for feature extraction, and then the selected feature sets were used for classification. The third problem was to quantify thermograms of multiple sclerosis (MS) patients for better assessment of the disease and monitoring the therapy. Thermal images of two patients and a healthy control from lower limbs were evaluated during experiments, and localized quantification of the effect of MS on the feet of the patients using thermal images method was proposed. The proposed method was fully correlated with the evaluations of physician. It is shown that medical thermal imaging has high potential in many fields of medicine as a non-invasive method for pre-diagnosis and follow-up.
Merkezi Olmayan Elektronik Sağlık Kaydı Yönetim Sistemi ve Makine Öğrenmesi Yöntemleri ile Hastalık Tahmini
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2024) Dedetürk, Beyhan Adanur; Güngör, Burcu
Electronic health records (EHRs) are vital to the advancement of healthcare and can help detect and prevent diseases early. However, EHR sharing faces challenges such as managing large data volumes, ensuring data privacy, security, and interoperability. This thesis aims to develop and analyze a blockchain-based EHR sharing system for disease prediction mechanism integration using SysML. The AguHyper platform, built by merging the InterPlanetary File System (IPFS) with Hyperledger Fabric, ensures the immutability of health records by storing hash values in the blockchain and encrypted records in IPFS. The system architecture and implementation configurations, including CouchDB and the Raft consensus mechanism, are thoroughly examined. The study also presents a novel hybrid approach called CSA-DE-LR, which integrates Differential Evolution (DE) and Clonal Selection Algorithm (CSA) with Logistic Regression (LR) to improve LR weights for precise categorization of cardiovascular diseases. The integration of the AguHyper with the CSA-DE-LR is explained in detail. At the end of our performance evaluations, we concluded that the AguHyper model has the potential to speed up the process of collecting and sharing data, and it offers an efficient platform for the participants.
MRG Taramalarında Alzheimer Hastalığının Zaman Dağılımlı Sınıflandırılması
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2024) Dündar, Mehmet Sait; Yılmaz, Bülent
This thesis presents a comprehensive framework for studying Alzheimer's Disease (AD) progression by focusing on the classification of AD, Mild Cognitive Impairment (MCI), and Cognitively Normal (CN) individuals using advanced machine learning models that analyze changes in brain volumetrics over time through MRI scans. In the first phase of the research, MR images from the Alzheimer's Disease Neuroimaging Initiative database were utilized, which included sequences of 3-4 scans taken annually from 22 CN, 18 AD, and 20 MCI subjects. Key volumetric parameters such as cortical thickness and intracranial volumes were extracted using the CAT12 toolbox in SPM software. A novel classification method based on the rate of volumetric changes over time was employed, effectively capturing the progressive nature of neurological changes. This approach achieved accuracies of 82.5% in distinguishing AD from CN, 71% in differentiating MCI from AD, and 69% in separating MCI from CN, alongside a 55% accuracy in a three-way classification using random forest and support vector machines. Building on these initial insights, the second phase of the study significantly advanced the methodology by integrating a pre-trained 3D ResNet 101 CNN algorithm for initial spatial categorization of MRI scans, followed by the use of Long Short-Term Memory (LSTM) networks. These LSTMs processed the same sequences of 3-4 annual scans for each patient, enhancing the model's ability to analyze and interpret the temporal progression of volumetric changes. This sophisticated approach led to marked improvements in classification accuracy: 96.7% in differentiating AD from CN, 87.5% in distinguishing AD from MCI, and 86.4% in separating MCI from CN. The study effectively demonstrates a significant enhancement in capturing the temporal dynamics of AD progression.
Optik Saçılma Temelli Rastgele Orman Destekli Parçacık Tespiti ve Sınıflandırılması
(Abdullah Gül Üniversitesi / Fen Bilimleri Enstitüsü, 2023) Genç, Sinan; Genç, Sinan; İçöz, Kutay; Erdem, Talha
Microplastics, tiny plastic particles with sizes smaller than 5 mm., are often found in oceans, rivers, lakes, and atmosphere due to plastic pollution. Microplastics releasing toxic chemicals threaten the environment and harm the aquatic life and humans. Especially, the accumulation of microplastics can have detrimental effects on the food chain as a result of larger organisms consuming smaller organisms. Detecting the microplastics is crucial but also challenging. Over the years, researchers have developed different detection methods. One of the standard methods is using spectroscopy tools such as Fourier transform infrared spectroscopy (FTIR) and Raman spectroscopy. These techniques can identify the chemical composition of microplastics, which can help determine their sources and potential impacts. Another method is the use of microscopy, which allows for the visualization and counting of microplastics in samples. However, these techniques require costly infrastructure, and these instruments being large in size significantly limits the mobility. As a remedy to the cost and mobility problems, in this thesis, we propose and demonstrate a low-cost, portable system to detect size, concentration, and refractive index of microplastics. Our system comprises of low-cost and low-weight components which are utilized for recording the scattering patterns of microplastics in aqueous media. We demonstrate successful predictions of the size and refractive index of microparticles at a given wavelength using a Random Forest Algorithm which relates the measured scattering pattern with the Mie theory. We further employ the refractive index information at various wavelengths for determining the material type of microplastics. We believe that our proposed system enabling an easy, fast, low-cost, and on-site detection of microplastics will be a beneficial tool for the fight against microplastics in the environment.
Protein Yapı Tahmini için Derin Öğrenme Modellerinin Geliştirilmesi
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2022) Görmez, Yasin; Aydın, Zafer
The three-dimensional structure of a protein provides important clues about the function of that protein. Although there have been many studies on protein structure prediction, the problem has still not been solved completely. As it is very difficult to predict the three-dimensional structure of a protein directly, predictions of structural properties of proteins such as secondary structure, solvent accessibility, and torsion angles are carried out first, which are later used as inputs to more elaborate structure estimation tasks. In this thesis, novel deep learning models have been developed by using convolutional neural networks (CNN), graph convolutional networks (GCN) and long-short-term memory (LSTM) recurrent neural networks to predict secondary structure, solvent accessibility and torsion angles of proteins. A rich feature set formed by using PSI-BLAST, HHBlits, physicochemical properties, structural profile matrices, AA index values, and graphs representing the relationship between amino acids were used as inputs to the models. In the first study, a deep learning model was developed by using CNN and GCN layers for secondary structure prediction. In the second study, LSTM layers were added to the first model, which was extended to make solvent accessibility and torsion angle predictions as well using the multi-task learning approach. In both studies, graphs were generated using neighborhood relations between amino acids. In the last study, a novel U-net-based model was designed for secondary structure prediction using CNN, GCN, and LSTM layers. The graph matrices used as input to GCN layers were obtained by using protein contact map prediction. All models were trained, optimized and tested on benchmark data sets. Improvements were obtained in accuracy as compared to the state-of-the-art
Su Altı Dalgıç Pompa Uygulamaları için Doğrudan Yol Vermeli Relüktans Motorunun Sistematik Olarak Tasarım Optimizasyonu ve Gerçeklemesi
(Abdullah Gül Üniversitesi, Fen Bilimleri Enstitüsü, 2022) Tekgün, Didem; Alan, İrfan
Considering the electric drive systems constitute roughly 40% of global energy production, improving electric machine efficiencies provides important nationwide and global scale advantages. Among the electric motors used in the industry, a major portion of them are pump motors used for pumping underground waters and petroleum products. Especially the motors for submersible pump applications run at very low-efficiency levels because of the motor design issues and wrong selection of motor-pump configurations. Due to the features like robustness, low cost, and line start capability, induction machines (IM) are generally the first choice for pump applications. However, IMs work with low efficiency, especially at low and medium power levels. Line start synchronous reluctance machines (LS-SynRM) come to the scene as a reasonable alternative by having the line start capability and not having rare earth permanent magnets as well. The working principle of these machines is a combination of a reluctance machine and an IM. In LS-SynRM, a rotor cage is inserted in the rotor for the machine to start with the line voltage, but the rotor copper losses become zero when the machine operates at synchronous speed. Moreover, SynRMs have higher power and torque density. In this thesis study, it is aimed to reduce the overall cost of the submersible water pump system by designing and optimizing a LS-SynRM as a submersible pump motor with higher efficiency compared to conventional IMs. Increasing the efficiency of the pump motor used in industry will improve the overall system performance. Accordingly, it lowers energy and maintenance costs, and easy process control will be achieved. This way, while reducing energy consumption nationwide significantly, not only the natural resources will be protected, but also huge amounts of money will be saved.

Browse

Browsing Doktora Tezleri by Department "Fen Bilimleri Enstitüsü / Elektrik ve Bilgisayar Mühendisliği Ana Bilim Dalı"