Bildiri Koleksiyonu
Bu koleksiyon için kalıcı URI
Güncel Gönderiler
Öğe Segmentation of astrocyte cells in fluorescently labeled images(2024) Pesen, Muhammed; Kayasandık, Cihan BilgeAstrocytes, the most abundant brain cells, play crucial roles beyond mere support for neurons. Despite their significance, understanding them remains challenging due to imaging and analysis limitations. In this study, we developed a segmentation model that improves upon existing methods and conducted a comprehensive comparison of approaches from the literature, predominantly focusing on various U-net architecture variants. By optimizing configurations, loss functions, and processing steps, our U-net with Inception layers achieved an 83% F1 score, excelling in densely populated regions. The effectiveness of the developed segmentation model was assessed by applying it to images of astrocyte cells with different morphologies taken from serum-containing and serum-free cultures. By applying the Directional Ratio method to the segmented cells, it was determined that astrocytes in serum-free cultures exhibited greater anisotropy, thus demonstrating how serum influences astrocyte morphology through computational methods. Our findings emphasize the importance of multi-scale analysis in cell segmentation. The proposed method effectively segments and analyzes astrocyte images, revealing significant morphological changes influenced by serum, and offers a valuable tool for neuroscience studies by minimizing manual errors and enabling large-scale analysis.Öğe Spatio-angular resolution trade-off in face recognition(2024) Alam, Muhammad Zeshan; Kelowani, Sousso; Elsaeidy, MohamedEnsuring robustness in face recognition systems across various challenging conditions is crucial for their versatility. State-of-the-art methods often incorporate additional information, such as depth, thermal, or angular data, to enhance performance. However, light field-based face recognition approaches that leverage angular information face computational limitations. This paper investigates the fundamental trade-off between spatio-angular resolution in light field representation to achieve improved face recognition performance. By utilizing macro-pixels with varying angular resolutions while maintaining the overall image size, we aim to quantify the impact of angular information at the expense of spatial resolution, while considering computational constraints. Our experimental results demonstrate a notable performance improvement in face recognition systems by increasing the angular resolution, up to a certain extent, at the cost of spatial resolution.Öğe Theoretical limits of soop tdoa localization of unmanned systems with imperfect synchronization(2023) Elgammal, Khaled Walid; Turan, Berke Can; Bedir, Oğuz; Çelebi, Hasari; Qaraqe, Marwa K.; Özdemir, Mehmet KemalSelf-localization is of crucial importance to Unmanned Aerial Vehicle (UAV) industry. Localization systems such as the Global Positioning System (GPS) are prone to jamming and spoofing. Localization using the Signal-Of-Opportunity (SoOP) gained much attention recently as an alternative to typical localization systems. We consider a general scenario where a receiver deployed on a mobile UAV finds its location based on Time Difference Of Arrival (TDOA) measurements with a number of unsynchronized reference Base Station (BS) receivers using signals from unsynchronized transmitters. The UAV has relative time offsets and clock skews with the BSs. We derive the Cramer-Rao Lower Bound (CRLB) of the UAV location and relative skews. Then, we plot them for Phase Alternating Line (PAL) analog TV transmission as SoOP.Öğe KoExPubMed: A tool for effective and customized knowledge extraction from PubMed(2023) Sailunaz, Kashfia; Jurca, Gabi; Beştepe, Deniz; Karatay, Büşra; Alhajj, Lama; Özyer, Tansel; Rokne, Jon; Alhajj, RedaAn exponential growth in the literature in general and the medical literature in particular raises a need for effective intelligent analysis strategies and tools to provide valuable insights to researchers about the current evolving literature. While existing applications provide more specific approaches to the problem, such as focusing on particular genome or protein information, in this paper, the proposed application provides effective and detailed analysis of PubMed. The developed tool, named KoExPubMed, follows a more generalized and holistic way by taking into consideration different types of information such as authors, countries, genes, and the interactions between them. The developed application consists of four main components; (1) keyword search and ID extraction, (2) PubMed article information and abstract retrieval, (3) country and address extraction, and (4) gene information extraction. In addition to the fundamental components, the tool provides a variety of visualization options for showing the extracted information and the related associations, including line charts for densities and countries, chord charts for collaborations of authors, network graphs for the genes mentioned together, bubble charts for gene frequencies, etc. By addressing the need for a generalized data mining tool, we propose a comprehensive application which is capable of employing data mining and machine learning techniques to extract from PubMed knowledge valuable to researchers and practitioners who are interested in closely investigating the achievements of others.Öğe Creating a learning profile by using face and emotion recognition(2023) Yurtdaş, Gözde; Alhajj, Loubaba; Sailunaz, Kashfia; Özyer, Tansel; Rokne, Jon; Alhajj, RedaThe aim of this work is to employ face recognition for creating learning profiles of the analysed persons who are students in this study. Generating education profiles will help experts in the diagnosis of Attention Deficit Hyperactivity Disorder (ADHD), which is a serious problem in children. Children with ADHD often have the ability and potential to learn. However, it may be difficult to reveal their capabilities and skills. Accordingly, a suffering child may have a hard time succeeding in real life when he/she is ignored and expected to mix with other children. The unrealized gap and deficiency may lead to other problems and more complicated situation with unpredictable consequences. Thanks to the system developed in this study, and the like, which will help in diagnosing the ADHD disease, and hence suffering individuals will be able to recognize their deficiencies, understand their ability to learn and adapt when approached differently in a way which suits his/her situation. This personalized handling of infected students will be an excellent guide to advance their potential and integration within the society carefully and smoothly. The system analyzes the face of a student to inspire his/her emotional state. The reported test results demonstrate how the system works well and produces high accuracy under a variety of severe conditions such as skewed angle, less illumination, accessories etc.Öğe Investigating the roles of micrornas / lncrnas in characterizing breast cancer subtypes and prognosis(2023) Pek, Reyhan Zeynep; Zavalsız, Muhammed Talha; Serdar, Melis; Alhajj, Lama; Alhajj, Sleiman; Sailunaz, Kashfia; Özyer, Tansel; Rokne, Jon; Alhajj, RedaMolecular subtyping is a method of separating tumor clusters in a cancer type with common features according to molecular data and classification models. Genome datasets are taken from many different people and some genetic material, more precisely genetic markers, are obtained to predict the presence of a disease. In addition, breast cancer occurs due to mutation or modification observed in cells. miRNAs and lncRNAs take participation in cell cycle, regulation, and even chromatic inhibition of cell. For example, miRNAs function in cell cycle regulation as the degradation of mRNAs. Therefore, the aim of this work is to investigate the roles of miRNAs and lncRNAs in prognosis and characterizing the subtypes of Breast Cancer.Öğe Exploring gene expression and clinical data for identifying prostate cancer severity levels using machine learning methods(Institute of Electrical and Electronics Engineers Inc., 2023) Marouf, Ahmed Al; Alhajj, Reda; Rokne, Jon G.; Ghose, Sunita; Bismar, Tarek A.Prostate cancer (PCa) is the most common type of cancer in men worldwide. It is a cancer that starts in the small walnut-shaped male gland called the prostate. From the prostate, it can form a metastasis into other organs. If detected and diagnosed early the survival rate may increase to 95%. Therefore, early detection and diagnosis are important tasks performed by a pathologist. The pathologist identifies the severity levels using a scale called the Gleason grading group (GGG). The GGG is found by pathologists by looking at a biopsy sample and assigning a grade of low, intermediate, or high to the sample. The pathologist then assesses a second sample in the same manner. The GGG is found by adding these two scores provides the total Gleason score. In this paper, we have explored tissue microarray (TMA) and clinical data collected by pathologists of Alberta Precision Laboratory, for predicting the severity level of prostate cancer using various machine learning methods. Traditional classifiers, such as Naïve Bayes, Decision Tree, Support Vector Machine with Radial basis function (RBF), Logistic Regression, and ensemble classifiers, such as Random Forest, and Bagging with k-nearest neighbors have been applied through the machine learning pipeline containing imputation and sampling techniques. An integrated SMOTE-Tomek Links method is adopted for handling the class imbalance problem. The highest accuracy obtained is 99.64% from the Random Forest method.Öğe Iterative kernel reconstruction for deep learning-based blind image super-resolution(IEEE Computer Society, 2022) Yıldırım, Süleyman; Ateş, Hasan Fehmi; Güntürk, Bahadır KürşatDeep learning based methods have received a great deal of interest in recent years to solve the single image superresolution (SISR) problem and their performance is proven to be superior when compared to classical SR techniques. Yet, most of these methods fail to generalize well on real life image datasets because they are trained on synthetic datasets with a small range of blur kernels. This makes data-driven approaches inherently weak when it comes to real images. Therefore, applying image super-resolution independently of the blur kernel is still a challenging task. In this paper we propose IKR-Net, Iterative Kernel Reconstruction network, for blind SISR. In the proposed approach, kernel estimation and high resolution image reconstruction are carried out iteratively using deep models. The iterative refinement provides significant improvement in both the reconstructed image and the estimated blur kernel. IKR-Net achieves state-of-the-art results in blind SISR, especially for images with motion blur.Öğe Swin transformer based siamese network for thermal and optical image registration(IEEE-Institute of Electrical and Electronics Engineers, 2023) Elsaeidy, Mohamed; Yağmur, İsmail Can; Ateş, Hasan Fehmi; Güntürk, Bahadır KürşatThe process of multi-modal image registration is fundamental in remote sensing and visual navigation applications. However, existing image registration methods that are designed for single modality images do not provide satisfactory results when applied to multi-modal image registration. In this research, our objective is to achieve highly accurate alignment of both infrared and optical (visible range) images. To accomplish this goal, we explore the effectiveness of the Swin Transformer encoder and cosine loss in enhancing the keypoint-based image registration process. Simulation results show the improvement achieved in multi-modal registration by using a transformer based Siamese network.Öğe Real-time web-based International Flight Tickets Recommendation System via Apache Spark(Institute of Electrical and Electronics Engineers Inc., 2023) Malkawi, Malek; Alhajj, RedaTraveling by airplane has become more popular with advanced technology. The tickets can be booked effortlessly via airlines corporation's online platforms. However, recommending the best airline ticket according to the buyer's demands is a challenging task owing to the unexpected fluctuations in the price depending on various reasons. Traditional recommender suggestions are optimized for predicting the price for a specific time or estimating the period of the lowest price. However, considering the sudden changes is an essential matter to increase the accuracy. In this work, we present a web-based real-time system to recommend the most suitable ticket regardless of the continuous changes in the prices. Apache Spark has been used to analyze the data obtained from the international airline web pages. Besides the ease of use of the system, it helps the customer to buy the flight ticket at the lowest price for the desired period and destination. Based on the proposed model, using Python programming language, Flask web server, and Apache Spark, we design and implement the international ticket recommendation system with the MVC design pattern.Öğe Classes versus communities: Outlier detection and removal in tabular datasets via social network analysis (ClaCO)(Institute of Electrical and Electronics Engineers Inc., 2023) Üçer, Serkan; Özyer, Tansel; Alhajj, RedaIn this research, we introduce a model to detect inconsistent & anomalous samples in tabular labeled datasets which are used in machine learning classification tasks, frequently. Our model, abbreviated as the ClaCO (Classes vs. Communities: SNA for Outlier Detection), first converts tabular data with labels into an attributed and labeled undirected network graph. Following the enrichment of the graph, it analyses the edge structure of the individual egonets, in terms of the class and community belongings, by introducing a new SNA metric named as 'the Consistency Score of a Node-CSoN'. Through an exhaustive analysis of the ego network of a node, CSoN tries to exhibit consistency of a node by examining the similarity of its immediate neighbors in terms of shared class and/or shared community belongings. To prove the efficiency of the proposed ClaCO, we employed it as a subsidiary method for detecting anomalous samples in the train part in the traditional ML classification task. With the help of this new consistency score, the least CSoN scored set of nodes flagged as outliers and removed from the training dataset, and remaining part fed into the ML model to see the effect on classification performance with the 'whole' dataset through competing outlier detection methods. We have shown this outlier detection model as an efficient method since it improves classification performance both on the whole dataset and reduced datasets with competing outlier detection methods, over several known both real-life and synthetic datasets.Öğe NetDriller-V3: A powerful social network analysis tool(Institute of Electrical and Electronics Engineers Inc., 2023) Afra, Salim; Özyer, Tansel; Rokne, Jon; Alhajj, RedaThe development in technology has led to the generation of huge amounts of data from various sources, including biological data, social networking data, etc. Accordingly, social network analysis has received considerable attention with the availability of more raw datasets which could be realized using a network structure. Most of the datasets can be represented as a social network which is a graph consisting of actors having relationships. Many tools exist for social network analysis inspired to extract knowledge from the networks. NetDriller has been developed as a social network extraction, manipulation and analysis tool to cover the lack that exists in other tools. It is capable of constructing social networks from raw data by employing a variety of data mining and machine learning techniques. In this paper, we describe an extend version of NetDriller, which has some new essential functions, including social network construction using data collection from Twitter, DBLP and IEEE. We also added (1) a new chart for viewing the network property and metrics, and (2) new graph manipulation techniques using GUI to keep the tool up to date with the huge volume of networks and the different types of raw data available on the web.Öğe A prediction approach for the functional effects of non-coding gene variants(Institute of Electrical and Electronics Engineers Inc., 2022) Yurtdaş, Gözde; Aslan, Kağan; Özyer, Sibel Tarıyan; Özyer, Tansel; Kaya, Mehmet; Alhajj, RedaThe aim of this study is to develop an approach for predicting the functional effects of variants of non-coding genes which have great importance in human genetics. Non-coding genes have formed a very vital field of study since they have a high effect on diseases. However, little is known about non-coding genes compared to coding genes, and they are found in the body almost 9 times more than coding genes. This is a critical issue, and i t is very important to predict the effects of these genes, which are so abundant in the body and difficult to understand. This exhibits the motivation of the study described in the paper. For this purpose, an extensive literature review was first conducted, and possible datasets that could be used were examined. Then, using Python programming language, we developed a prediction model with high accuracy. After investigating how important non-coding gene variants are, and in what areas they can be used, we decided to use a functional interaction network from the deep learning models as the most suitable method. We used STRING (Search Tool for the Retrieval of Interacting Genes/Proteins) which is a biological database and web resource of known and predicted protein-protein interactions. As a second step, we generated feature vectors. After checking the overlap of non-coding genes, we extracted three types of feature vectors. Identifying protein interaction network in Python, the outcome describes the interplay between the biomolecules encoded by genes. It allows to understand the complexities of cellular functions, and even predict potential therapeutics. As a last step, we implemented a deep learning model which included three fully connected (FC) layers, also known as dense layers, with dimensions 40, 10, and 2, respectively. Experimental results demonstrate that the proposed method captured high accuracy values.Öğe Pancreatic tumor detection by convolutional neural networks(Institute of Electrical and Electronics Engineers Inc., 2022) Zavalsız, Muhammed Talha; Alhajj, Sleiman; Sailunaz, Kashfia; Özyer, Tansel; Alhajj, RedaArtificial Intelligence and its sub-branches like Machine Learning (ML) and Deep Learning (DL) applications have the potential to have positive effects that can directly affect human life. Medical imaging provides a way for the internal structure of the human body to be visible with various methods. With DL models, cancer detection, which is one of the most lethal diseases in the world, from medical images can be made possible with high accuracy. The main objective of this paper is to detect Pancreatic Cancer, which is one of the cancer types with the highest fatality rate, from a dataset of Computed Tomography (CT) images, which is one of the medical imaging techniques and has an effective structure in Pancreatic Cancer imaging. The designed DL model is integrated into the Flask application to develop a web application. With this application, early diagnosis of pancreatic cancer can be achieved, which progresses insidiously and therefore does not spread to neighboring tissues and organs when the treatment process is started. Due to the abundance of medical images reviewed by medical professionals, this application can assist radiologists and other specialists in Pancreatic Tumor detection.Öğe Hybrid CPU-GPU acceleration of a multithreaded image stitching algorithm(Institute of Electrical and Electronics Engineers Inc., 2022) Tesfay, Shewit W.; Demirdağ, Zeynep Gülbeyaz; Uğurdağ, H. Fatih; Ateş, Hasan FehmiReal-time image stitching is critical, especially in un-manned aerial vehicles, and its acceleration has received attention in recent years. This paper describes an image stitching acceleration scheme for heterogeneous (CPU+GPU) devices. Acceleration is attempted with both multithreading and multiprocessing. Most time-critical functions in the algorithm are offloaded on to the GPU. We crafted a 3-buffer ping-pong mechanism for synchro-nization and data transfer among threads/processes in order to maximize CPU utilization. We carried out our experiments on Nvidia Jetson AGX Xavier. Results show that more than 3x acceleration is achieved.Öğe Infrared-to-optical image translation for keypoint-based image registration(Institute of Electrical and Electronics Engineers Inc., 2022) Elsaeidy, Mohamed; Erkol, Muhammed Emin; Güntürk, Bahadır Kürşat; Ateş, Hasan FehmiMulti-modal image registration is a critical step in many remote sensing and visual navigation applications. While image registration techniques developed for single modality images do not perform well for multi-modal registration, techniques developed for multi-modal image registration are not suitable for real-time applications. In this study, we aim to achieve real-time registration of infrared and optical (visible range) images. Specifically, we investigate the use of image-to-image translation deep network to convert infrared images to optical images, which is then applied for keypoint-based image registration.Öğe rs-fMRI analysis using spatio-temporal sparse convolutional neural networks(Institute of Electrical and Electronics Engineers Inc., 2022) Yener, Fatma Muberr; Yıldız, Sultan; Hafeez, Muhammad Adeel; Kayasandık, Cihan Bilge; Doğan, Merve YüsraNeuropsychiatric diseases such as Autism Spectrum Disorder (ASD) and Schizophrenia cause various behavioral and communication dysfunctions in human life. Resting state functional magnetic resonance imaging (rs-fMRI) is used to detect and characterize functional changes in the brain associated with these disorders. Machine learning methods are known to perform well in classifying fMRI images and have proven to have great potential in the field of computer aided diagnosis. In most of the previous studies, hand-crafted features have been used in fMRI analyzes and classifications to date. This prevents the system from being end-to-end and causes spatial or temporal information to be lost due to dimension reduction. The method presented in this study works end-to-end as well as being fed with an entire 4-dimensional fMRI sequence. It is faster than traditional convolutions and recurrent neural networks of the same size, thanks to the sparse convolutional layers that are the building blocks of the network. Experiments with schizophrenia and ASD fMRIs have shown similar performance to those in the literature, despite limited resources.Öğe Automated analysis of the EEG signals for prediction of possible effectiveness of rTMS treatment in alzheimer's patient(Institute of Electrical and Electronics Engineers Inc., 2022) Duzman, Hacer; Torlak, Meryem; Hindi, Osama Ali; Kayasandık, Cihan BilgeAlzheimer's disease (AD) is a progressive, chronic neurodegenerative brain disease that generally infects the elderly. The analysis of electroencephalography (EEG) signals has been commonly used for diagnosis. Repetitive transcranial magnetic stimulation (rTMS) is one of the most significant non-pharmacological methods that offer a potential treatment for neurological and psychiatric diseases. Nevertheless, recent studies have shown that patients do not benefit from this treatment at the same level. In that case, there would be a loss of time, money, and effort in the application of treatment. In this project, the EEG data is collected and then analyzed through multichannel analysis using various machine-learning methods. There are 14 patient datasets available for analysis. Thus, the main aim is to find the most significant features of the patients' EEG signals who were treated by rTMS. The novelty of this project lies in performing multichannel analysis and finding a personalized treatment for AD by combining both EEG Analysis and rTMS treatment; where such a combination has not been done yet in any project before. Even in the studies that obtained good accuracy results, there was a lot of useful information missed due to the absence of multichannel analysis. As with multichannel analysis, the data of each channel is analyzed separately. The results showed that the benefit of rTMS treatment can be distinguished for each patient based mainly on the following features of their EEG data signals: Alpha, Beta, and Delta band power features, besides the complexity feature.Öğe Brain tumor classification using MRI images and convolutional neural networks(Institute of Electrical and Electronics Engineers Inc., 2022) Hafeez, Muhammad Adeel; Kayasandık, Cihan Bilge; Doğan, Merve YüşraThe brain tumor has become one of the most prominent types of cancers affecting a huge population across the globe every year. It has the lowest life expectancy rate and the risk of death is highly associated with the type, shape, and location of the tumor. The Magnetic Resonance Imaging (MRI) is a strong tool to detect different brain lesions and is extensively used by radiologists and physicians. For the early and accurate diagnosis of the brain tumor using MRI, it is important to consider automated computer-assisted diagnosis which is more flexible and efficient. In this paper, we have proposed a Convolutional Neural Network (CNN) based approach for the classification of three types of brain tumors (meningiomas, gliomas, and pituitary tumors). A publicly available dataset that contains 3064 T1-weighted brain CE-MRI images collected from 233 patients has been used in the study. We propose a 15 layers CNN model for the classification of three types of brain tumors from the mentioned dataset. We obtained an accuracy, precision, recall, and f1-score of 98.6%, 99%, 98.3%, and 98.6% from our proposed model which is higher than previously reported results.Öğe Convex hull in brain tumor segmentation(Springer Science and Business Media Deutschland GmbH, 2022) Sailunaz, Kashfia; Beştepe, Deniz; Alhajj, Sleiman; Özyer, Tansel; Rokne, Jon; Alhajj, RedaTumors are the second leading cause of death. Among the tumors, brain tumors constitute one of the most complex tumor categories with a high mortality rate. Therefore, brain tumor detection and segmentation from non-invasive imaging like MRI is an important research area. Although most recent researches for brain tumor detection are focused on deep learning methods, machine learning, geometrical approaches, thresholding and hybrid models are also explored frequently. In this paper, a novel brain tumor segmentation method containing thresholding, computational geometry and heuristics is proposed. The proposed model is tested with two brain tumor datasets to show comparative results for brain tumor segmentation with thresholding, convex hull and an area heuristic. The application of different filtering on a direct convex hull model and a heuristic-based convex hull model shows that the convex area based heuristic with the convex hull approach is able to segment brain tumors more accurately than previous approaches.
- «
- 1 (current)
- 2
- 3
- »
