Gümüşhane ÜniversitesiGümüşhane ÜniversitesiAçık Erişim Arşivi · RDL Open

Sürdürülebilir Kalkınma Amaçları

Yazarlar

Yayınlar

Yeniden eskiyeEskiden yeniye
  • BildiriAçık Erişim2024
    An Efficient Technique Based on Deep Learning for Automatic Focusing in Microscopic System

    Dogu, Fatma Tuana; Dogan, Hulya; Dogan, Ramazan Ozgur; Ay, Ilyas; Sezen, Sena F.; Seyman, M.

    During microscopic examination process, experts firstly determine the most appropriate focusing point by precisely moving the microscope stage in the Z axis. This manual focusing process may cause variability in results depending on hand-eye coordination and individual experience. Therefore, the development of automatic focusing offers superior quality and consistent imaging. Many researchers have studied automatic focusing systems and proposed many focus functions to extract focus information from multi-focus images. Nevertheless, these focusing systems still consist of several significant limitations such as requiring a supplementary material, longer running time, producing different performance depending on the sample and magnification objective. In this study, an efficient technique based on deep learning is proposed to automatically determine the focused image during microscopic examination processes to minimize these limitations in the literature. This technique takes a sequence of images with the same field of view and different focuses as input and gives the image with maximum focusing as output. The proposed technique is compared with other automatic focusing techniques on the literature. In this study, novel multi-focus image sequences obtained from liver, intestine, heart, kidney, and lung samples is prepared to evaluate the performance of automatic focusing techniques. These sequences are obtained by scanning with 10× and 40× magnification objectives. Running time, accuracy, number of local maximum points, range, and noise levels are used as evaluation criteria in this study. According to evaluation criteria results obtained in this study, it has been proven that the proposed technique provides better performance than other automatic focusing techniques.

    İncele
  • MakaleAçık Erişim2024
    A study on effective data preprocessing and augmentation method in diabetic retinopathy classification using pre-trained deep learning approaches

    İncir, Ramazan; Bozkurt, Ferhat

    High glucose levels in the blood not only damage different tissues and organs of the body, but also cause adverse effects on the eye. This condition is called diabetic retinopathy (DR). DR can cause blurred vision, darkening of the field of vision, and severe vision loss. The number of people infected with the disease is increasing in our country and worldwide. The time-consuming physician check-ups and the presence of small lesions indicate the need to develop diagnostic systems. Deep learning-based applications have become the trend for diagnosing and grading diseases from images. This study aims to create a meaningful and sufficient dataset using effective data preprocessing and affine transformation techniques in diabetic retinopathy classification. In this study, classification was performed using seven different pre-trained deep learning architectures. An experimental study of each technique was performed on the EyePACS dataset. An overfitting problem was encountered in the experimental results with the original data set. Thus, data preprocessing and data augmentation processes were carried out in order to eliminate overfitting by considering the imbalance between classes in the dataset. The classification performance obtained from each architecture was observed according to performance metrics of precision, recall, F1 Score, accuracy, and loss. In this study, the best performance was achieved with 97.65% test accuracy with the proposed EfficientNetV2-M network model.

    • SDG 3 · Sağlık ve Kaliteli Yaşam
    İncele
  • MakaleAçık Erişim2024
    Optimizing Memristor-Based Synaptic Devices for Enhanced Energy Efficiency and Accuracy in Neuromorphic Machine Learning

    Gökgöz, Baki; Aydin, Tolga; Gül, Fatih

    The traditional Von Neumann computing architecture, which necessitates data transfer between external memory and the processor, incurs significant energy and time costs when running deep learning (DL) and machine learning (ML) architectures. The primary issue with the energy and time efficiency of this architecture stems from the frequent and intensive data transfers between memory and the processor. Therefore, memristive synaptic devices are utilized to overcome this energy and time inefficiency while performing cognitive tasks. The fundamental working principle of memristive devices is to reduce the need for data transfer by combining memory and processing in the same location, thereby significantly decreasing both energy consumption and the time required for operations. However, to achieve the desired level of efficiency in terms of energy and time consumption from neuromorphic systems, the performance of these systems needs to be further improved with respect to accuracy and test error rates for classification applications. Achieving high accuracy performance in such deep learning or machine learning models necessitates optimization processes not only at the hardware level but also at the algorithmic level. In this context, this paper presents a comprehensive examination and comparison of the frequently used SGD and its momentum variants for deep learning and machine learning applications in memristor-based neuromorphic computing systems. The study thoroughly investigates the performance of critical metrics such as the learning properties, energy efficiency, and accuracy rates of the nano-scale titanium dioxide (TiO2) based synaptic device. The experimental results for the MNIST dataset showed AdaDelta 89.48%, AdaGrad 79.00%, Adam 79.13%, AdaMax 79.68%, Momentum 88.55%, Nadam 81.20%, RMSprop 84.91% and SGD 89.47% accuracy. The experimental results for the CIFAR dataset showed AdaDelta 90.51%, AdaGrad 82.08%, Adam 83.10%, AdaMax 81.76%, Momentum 91.25%, Nadam 82.45%, RMSprop 88.11% and SGD 90.21% accuracy. 2024 The Authors. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.

    İncele
  • MakaleAçık Erişim2024
    Enhancing brain tumor classification through ensemble attention mechanism

    CELIK, Fatih; CELIK, Kemal; CELIK, Ayse

    Brain tumors pose a serious threat to public health, impacting thousands of individuals directly or indirectly worldwide. Timely and accurate detection of these tumors is crucial for effective treatment and enhancing the quality of patients’ lives. The widely used brain imaging technique is magnetic resonance imaging, the precise identification of brain tumors in MRI images is challenging due to the diverse anatomical structures. This paper introduces an innovative approach known as the ensemble attention mechanism to address this challenge. Initially, the approach uses two networks to extract intermediate- and final-level feature maps from MobileNetV3 and EfficientNetB7. This assists in gathering the relevant feature maps from the different models at different levels. Then, the technique incorporates a co-attention mechanism into the intermediate and final feature map levels on both networks and ensembles them. This directs attention to certain regions to extract global-level features at different levels. Ensemble of attentive feature maps enabling the precise detection of various feature patterns within brain tumor images at both model, local, and global levels. This leads to an improvement in the classification process. The proposed system was evaluated on the Figshare dataset and achieved an accuracy of 98.94%, and 98.48% for the BraTS 2019 dataset which is superior to other methods. Thus, it is robust and suitable for brain tumor detection in healthcare systems.

    • SDG 3 · Sağlık ve Kaliteli Yaşam
    İncele
  • Derleme MakaleYayıncıda açık erişim2024
    An overview memristor based hardware accelerators for deep neural network

    Gökgöz, Baki; Gül, Fatih; Aydın, Tolga

    The prevalence of artificial intelligence applications using artificial neural network architectures for functions such as natural language processing, text prediction, object detection, speech, and image recognition has significantly increased in today's world. The computational functions performed by artificial neural networks in classical applications require intensive and large-scale data movement between memory and processing units. Various software and hardware efforts are being made to perform these operations more efficiently. Despite these efforts, latency in data traffic and the substantial amount of energy consumed in data processing emerge as bottleneck disadvantages of the Von Neumann architecture. To overcome this bottleneck problem, it is necessary to develop hardware units specific to artificial intelligence applications. For this purpose, neuro-inspired computing chips are believed to provide an effective approach by designing and integrating a set of features inspired by neurobiological systems at the hardware level to address the problems arising in artificial intelligence applications. The most notable among these approaches is memristor-based neuromorphic computing systems. Memristors are seen as promising devices for hardware-level improvement in terms of speed and energy because they possess non-volatile memory and exhibit analog behavior. They enable effective storage and processing of synaptic weights, offering solutions for hardware-level development. Taking into account these advantages of memristors, this study examines the research conducted on artificial neural networks and hardware that can directly perform deep learning functions and mimic the biological brain, which is different from classical systems in today's context.

    İncele
  • BildiriYayıncıda açık erişim2024
    High Performance Automatic ROI Extraction from Hand Images with Convolutional Neural Networks

    Kosmaz Sunnetci, Betul; Bingol, Ozkan; Gedikli, Eyup; Ekinci, Murat

    Automatic ROI extraction from hand images is a critical step in palmprint recognition. Traditional methods involve steps such as identifying the region of the hand, finding the valleys between the fingers and marking the palm area. Some of these processes are performed manually and some are performed with image processing techniques. However, these methods are time-consuming, vulnerable to human error, and computationally redundant. In this study, a deep learning-based convolutional neural network is developed for high-performance automatic ROI extraction from hand images. The proposed method eliminates the hand segmentation process and provides a solution where the palm region can be automatically selected directly. This reduces processing time, minimizes human error and improves overall accuracy.

    İncele
  • Yüksek Lisans TeziAçık Erişim2023
    Evrişimsel sinir ağı kullanılarak göğüs radyografilerinde yüksek başarımlı bölütleme yapılması ve pnömoni tespitine etkisinin incelenmesi

    İlhan Aydın

    Zatürre diğer adıyla pnömoni bakteriyel veya viral enfeksiyonun neden olduğu yaşamı tehdit eden akciğer hastalığıdır. Bu nedenle, hastalığın erken teşhisi ölüm oranını önemli derecede azaltmaktadır. Göğüs röntgenleri (radyografi) hastalığın teşhisinde büyük öneme sahiptir. Pnömoni teşhisinde sağlık alanındaki çalışmalar dışında, mühendislik alanında da birçok çalışma yapılmakta ve teşhis sürecini hızlandırma konusunda kolaylıklar sağlamaktadır. Bu uygulamalardan biri de derin öğrenme yöntemleri ile hastalığı teşhis etmektir. Çalışmada, Python programlama dilinde Tensorflow Kütüphanesi ve erişime açık veri setlerinden yararlanılmıştır. Veri seti olarak içinde 1-5 yaş arası çocukların göğüs röntgenlerinin bulunduğu Veriseti-1 ve bu veri setinden bir kısım görüntülerin işaretli verilerinin de bulunduğu Veriseti-2 kullanılmıştır. İki aşamalı evrişimsel sinir ağı modeli oluşturulmuştur. İlk aşamada, akciğer bölgesi işaretlenmiş Veriseti-2 üzerinde değiştirilmiş U-Net mimarisi kullanılarak bölütlemesi yapılacak Veriseti-1 görüntüleri için model oluşturulmuştur. Ayrıca çalışmanın son kısmında göğüs kafesi bölgesini tespit edebilecek bir yaklaşım önerilmiştir. Oluşturulan model Veriseti-1'de kullanılarak bölütleme yapılmıştır. Bölütleme doğruluğu (accuracy) % 98.98 olarak hesaplanmıştır. Tasarlanan evrişimsel sinir ağı modeli Veriseti-2'ye uygulanmış ve pnömoni tespitinde test doğruluk (accuracy) değeri %98.55 olarak elde edilmiştir.

    İncele
  • BildiriYalnız künye2023
    Deep Learning Based Focus Measurement Operator for 3D Imaging in Microscopic Systems

    Dogan, Hulya; Dogan, Ramazan Ozgur

    Since 3D imaging in microscopic systems makes cell and tissue structures more prominent and visible, it provides higher performance in image processing applications such as automatic segmentation, recognition, classification, alignment and stitching. For this reason, 3D imaging in microscopic systems has been one of the most comprehensive and long-studied research areas in the literature. It seems that the most frequently preferred strategy for 3D imaging is shape from focus. Shape from focus is a strategy that uses a sequence of 2D images with the same perspective and different focused regions to create the 3D shape of the sample. Focus measurement operators are applied to extract the focus properties of the input images in shape from focus strategies. These operators have several limitations, such as more precise characterization of curves and edges in images. Researchers often use a pre- or post-processing technique to minimize these limitations, rather than proposing a new focus measurement operator. In this study, a deep learning based focus measurement operator, which does not require a pre- or post-processing technique, is developed to eliminate these limitations. The qualitative and quantitative results obtained in the study show that the proposed focus measurement operator is more effective in 3D imaging than other operators in the literature.

    İncele
  • BildiriYalnız künye2023
    An Extensive Study: Creation of A New Inverted Microscope Image Data Set and Improving Auto-Encoder Models for Higher Accuracy Segmentation of HaCaT Cell Culture Line

    Sevim, Zeynep; Dogan, Hulya; Demir, Zeynep; Sezen, Feride Sena; Dogan, Ramazan Ozgur

    In cell culture investigations, it is assumed that the number of cells is counted automatically without the need for forward planning, additional tools, or consumables. The automatic cell counting process is greatly facilitated and made functional by the segmentation of cell culture lines. Many studies with high-accuracy results have been developed for cell or tissue segmentation in images acquired from microscopic systems, but this success cannot be achieved for segmentation of cell culture line (especially in images acquired from inverted microscope). There are still restrictions in segmentation of cell culture line studies in inverted microscope systems. In this context, a new inverted microscope image data set is created and higher accuracy auto-encoder models are developed for segmentation of HaCat cell culture line. The analyzes of the developed auto-encoder models are evaluated using four different performance evaluation procedures which are Accuracy, Sensitivity, Specificity and Jaccard. In terms of the performance evaluation procedures used in this study, the achievements of the auto-encoder models are clearly satisfactory for the automated segmentation of the HaCat cell culture line.

    İncele
  • BildiriYalnız künye2023
    Pedestrian Inertial Navigation with Multi-Head CNN

    Cetin, Gokhan; Kucuk, Mehmet Ali; Koroglu, Muhammed Taha

    Traditional pedestrian dead-reckoning (DR) methods utilizing inertial measurement units (IMU) have been successfully employed in GPS-denied environments over the past two decades. However, despite continuous advancements in wearable IMU technology, the long-term solution still suffers from the issue of drift. In light of this, Deep Learning (DL) has emerged as a potential candidate for inertial navigation solutions, leveraging its powerful non-linear regression capabilities. Recent research has demonstrated that DL models trained on public datasets can effectively serve as end-to-end solutions. Most studies focus on smartphones, where the IMU is typically situated at the torso, and generally work with transformed data represented in the navigation coordinate frame to estimate pedestrian velocity and heading change at rates lower than the IMU sampling rate. In contrast, this study uses raw foot-mounted IMU data as input to a multi-headed structure consisting of independent convolutional neural networks (CNNs) and targets displacement directly and heading change. The proposed model is trained on a publicly available foot-mounted IMU dataset where pseudo ground-truth (GT) data is generated heuristically to provide supervision for the learning process. The results indicate that the Multi-Head CNN (MHCNN) model is capable of producing pedestrian trajectories at the IMU sampling rate, with positioning errors of less than 2m throughout the duration of the walks. The code for the trained network can be accessed in the associated repository, i.e., github.com/mtkoroglu/PIN-MHCNN.

    İncele
  • MakaleAçık Erişim2023
    Modelling some stand parameters using Landsat 8 OLI and Sentinel-2 satellite images by machine learning techniques: a case study in Türkiye

    Bulut, Sinan; Günlü, Alkan; Çakır, Günay

    Remote sensing technologies have been extensively used in forest management in predicting stand parameters. The goal of this study is to use Landsat 8 and Sentinel-2 satellite images to estimate stand volume, basal area, number of trees, mean diameter, and top height. 180 temporary sample plots were taken in pure Crimean pine stands with varied structure. Reflectance, vegetation indices, and eight texture values were generated from Landsat 8 and Sentinel-2 satellite images. The stand parameters were modelled with the remotely sensed data using multiple linear regression, support vector machine, and deep learning techniques. The results showed that the support vector machine technique provided the highest level of model performance with 45° orientation for number of trees (R2 = 0.98, RMSE%=5.97) and 90° orientation for basal area (R2=0.91, RMSE%=15.22). The results indicated that the texture values presented better results than the reflectance and the vegetation indices in estimating the stand parameters.

    • SDG 15 · Karasal Yaşam
    İncele
  • MakaleYalnız künye2023
    Detection of forest fire using deep convolutional neural networks with transfer learning approach[Formula presented]

    Reis, Hatice Catal; Turk, Veysel

    Forest fires caused by natural causes such as climate change, temperature increase, lightning strikes, volcanic activity or human effects are among the world's most dangerous, deadly, and destructive disasters. Detection, prevention, and extinguishing forest fires is challenging. In addition, forest fires can cause habitat destruction that cannot be controlled in time and cause great material and moral losses. Therefore, fast and accurate Detection of forest fires is vital in emergency response. Here, in solving the problem, the transfer learning method from deep learning sub-topics can be used, which allows the application of pre-trained networks to a new problem. The Fire Luminosity Airborne-based Machine learning Evaluation dataset (consisting of forest fire images) obtained by Unmanned Aerial Vehicle was used in this study. In the Detection of forest fire images in the dataset, InceptionV3, DenseNet121, ResNet50V2, NASNetMobile, VGG-19 deep learning algorithms, transfer learning techniques that can produce more successful results than networks trained from scratch, and hybrid proposed with Support Vector Machine, Random Forest, Bidirectional Long Short-Term Memory, Gated Recurrent Unit algorithms methods have been applied. In the classification study with the Fire Luminosity Airborne-based Machine learning Evaluation dataset in performance measurement, 97.95% accuracy was obtained from the DenseNet121 model, which was started with random weights. In the transfer learning study using ImageNet weights, satisfactory results were obtained with 99.32% accuracy in the DenseNet121 model. We anticipate that working in forest fire detection and response can be entirely satisfactory.

    • SDG 15 · Karasal Yaşam
    İncele
  • BildiriYalnız künye2022
    Segmentation of Pectoral Muscle Region in MLO Mammography Images by Backboned U-Net

    Dogan, Ramazan Ozgur; Türe, Hayati; Kayikçioglu, Temel

    The pectoral muscle region on MLO mammography images appears prominently similar to suspicious areas. For this reason, Computer-Aided Detection (CAD) systems remove this region to reduce false-positive rates in the mass detection process. In some cases, the pectoral muscle region is exposed to distortions due to the superposition effects caused by the mammography technique. As a result, segmentation error rates of the pectoral muscle region, whose characteristic features are deteriorated, appear. In this study, a method to identify impaired pectoral muscle regions with MobileNetV2 backboned U-Net Deep Learning method is proposed. The proposed method was tested on 84 and 201 mammography images taken from both MIAS and InBreast databases and segmented with 1.81% and 1.92% false-negative (FN) and 0.25% and 0.37% false positive (FP) rates, respectively. Particularly for distorted pectoral muscle regions, the proposed method has been shown to outperform some pioneering studies in this area.

    • SDG 3 · Sağlık ve Kaliteli Yaşam
    İncele
  • MakaleAçık Erişim2022
    COVID-19 diagnosis with deep learning; [Diagnóstico de COVID-19 con Deep Learning]

    Reis, Hatice Catal

    The coronavirus disease 2019 (COVID-19) is fatal and spreading rapidly. Early detection and diagnosis of the COVID-19 infection will prevent rapid spread. This study aims to automatically detect COVID-19 through a chest computed tomography (CT) dataset. The standard models for automatic COVID-19 detection using raw chest CT images are presented. This study uses convolutional neural network (CNN), Zeiler and Fergus network (ZFNet), and dense convolutional network-121 (DenseNet121) architectures of deep convolutional neural network models. The proposed models are presented to provide accurate diagnosis for binary classification. The datasets were obtained from a public database. This retrospective study included 757 chest CT images (360 confirmed COVID-19 and 397 non-COVID-19 chest CT images). The algorithms were coded using the Python programming language. The performance metrics used were accuracy, precision, recall, F1-score, and ROC-AUC. Comparative analyses are presented between the three models by considering hyper-parameter factors to find the best model. We obtained the best performance, with an accuracy of 94,7%, a recall of 90%, a precision of 100%, and an F1-score of 94,7% from the CNN model. As a result, the CNN algorithm is more accurate and precise than the ZFNet and DenseNet121 models. This study can present a second point of view to medical staff.

    • SDG 3 · Sağlık ve Kaliteli Yaşam
    İncele
  • MakaleAçık Erişim2022
    Comparison of deep learning and regression-based MPPT algorithms in PV systems

    Karabi̇Naoğlu, Murat Salim; Çakir, Bekir; Başoğlu, Mustafa Engin; Kazdaloğlu, Abdülvehhab; Güneroğlu, Aziz

    Solar energy systems (SES) and photovoltaic (PV) modules should be operated at the maximum power point (MPP) to achieve the highest efficiency in the energy generation processes. Maximum power point tracking (MPPT) applications using conventional methods may not be able to follow the global MPP (GMPP) of the PV system under changing atmospheric conditions and they could oscillate around the local MPP. In this study, a machine learning and deep learning (DL) based long short-term memory (LSTM) model is proposed as an innovative solution for MPPT. Contrary to the traditional MPPT applications using current and voltage sensors, the output resistance of the PV module estimation was made by using environmental parameters (such as temperature and radiation) and artificial intelligence algorithms in this study.The LSTM model was compared with artificial neural networks (ANN) and regression methods regarding mean square error (MSE), root mean square error(RMSE) and mean absolute error (MAE) parameters. It has been determined that the LSTM model has a better performance and could more successfully follow MPP compared to the other methods. Finally, after the comparison with the ANN method, it is proved that LSTM gives 37%, 21%, and 31% more successful MSE, RMSE, and MAE results, respectively.

    • SDG 7 · Erişilebilir ve Temiz Enerji
    • SDG 9 · Sanayi, Yenilikçilik ve Altyapı
    • SDG 13 · İklim Eylemi
    İncele
  • Derleme MakaleAçık Erişim2022
    Suspicious activity recognition for monitoring cheating in exams

    Genemo, Musa Dima

    Video processing is getting special attention from research and industries. The existence of smart surveillance cameras with high processing power has opened the for making it conceivable to design intelligent visual surveillance systems. Know a day it is very possible to assure invigilators safety during the examination period. This work aims to distinguish the suspicious activities of students during the exam for surveillance examination halls. For this, a 63 layers deep CNN model is suggested and named "L4-BranchedActionNet". The suggested CNN structure is centered on the alteration of VGG-16 with added four blanched. The developed framework is initially turned into a pre-trained framework by using the SoftMax function to train it on the CUI-EXAM dataset. The dataset for detecting suspicious activity is subsequently sent to this pre-trained algorithm for feature extraction. Feature subset optimization is applied to the deep features that have been obtained. These extracted features are first entropy coded, and then an ant colony system (ACS) is used to optimize the entropy-based coded features. The configured features are then input into a variety of classification models based on SVM and KNN. With a performance of 0.9299 in terms of accuracy, the cubic SVM gets the greatest efficiency scores. The suggested model was further tested on the CIFAR-100 dataset, and it was shown to be accurate to the tune of 0.89796. The result indicates the suggested frameworks soundness.

    • SDG 16 · Barış, Adalet ve Güçlü Kurumlar
    İncele
  • MakaleAçık Erişim2022
    Automatic prediction of covid-19 from chest-computed tomography (CT) images using deep learning architectures; [Derin öğrenme mimarilerini kullanarak göğüs BT görüntülerinden otomatik Covid-19 tahmini]

    Turk, Veysel; Catal Reis, Hatice; Kaya, Serhat

    Machine learning has been actively used in disease detection and segmentation in recent years. For the last few years, the world has been coping with the Coronavirus disease 2019 (COVID-19) pandemic. Chest-computerized tomography (CT) is often a meaningful way to detect and detect patients with possible COVID-19. This study aims to classify COVID-19 and non-COVID-19 chest-CT images using deep learning (DL) algorithms and investigate whether we can achieve successful results in different parameters using four architectures. The study was performed on proved positive COVID-19 CT images, and the datasets were obtained from the GitHub public platform. The study evaluated four different deep learning architectures of VGG16, VGG19, LeNet-5, and MobileNet. The performance evaluations were used with ROC curve, recall, accuracy, F1-score, precision, and Root Mean Square Error (RMSE). MobileNet model showed the best result ; F1 score of 95%, the accuracy of 95%, the precision of 100%, recall of 90%, AUC of 95%, and RMSE of 0.23. On the other hand, VGG 19 model gave the lowest performance; F1 score of 90%, the accuracy of 89%, the precision of 90%, recall of 90%, AUC of 89%, and RMSE of 0.32. When the algorithms' performances were compared, the highest accuracy was obtained from MobileNet, LeNet-5, VGG16, and VGG19, respectively. This study has proven the usefulness of deep learning models to detect COVID-19 in chest-CT images based on the proposed model framework. Therefore, it can contribute to the literature in Medical and Engineering in COVID-19 detection research.

    • SDG 3 · Sağlık ve Kaliteli Yaşam
    İncele
  • BildiriAçık Erişim2021
    Pneumonia Detection from Pediatric Lung X-Ray Images with Convolutional Neural Network Method

    Demir, Yasin; Bingöl, Ozkan

    Pneumonia can be cause diseases such as cough, fever and respiratory problems and to death in the progressive situations. Mostly lung x-ray images are used for diagnosis of pneumonia. In this study, an unique convolutional neural network model that can help experts in detection of pneumonia using lung x-ray images below 5 years of age. In the model, open access to 5840 images has been used, and various data increasing methods were used to increase the performance of the training and the success of the method. With the proposed convolutional neural network model, an accuracy rate of 94.55% has been achieved.

    • SDG 3 · Sağlık ve Kaliteli Yaşam
    İncele
  • MakaleAçık Erişim2021
    A Two-Phase Approach using Mask R-CNN and 3D U-Net for High-Accuracy Automatic Segmentation of Pancreas in CT Imaging

    Dogan, Ramazan Ozgur; Dogan, Hulya; Bayrak, Coskun; Kayikcioglu, Temel

    Background and objective: The size, shape, and position of the pancreas are affected by the patient characteristics such as age, sex, adiposity. Owing to more complex anatomical structures (size, shape, and position) of the pancreas, specialists have some difficulties in the analysis of pancreatic diseases (diabetes, pancreatic cancer, pancreatitis). Therefore, the treatment of the disease requires enormous time and depends on the experience of specialists. In order to decrease the rate of pancreatic disease deaths and to assist the specialist in the analysis of pancreatic diseases, automatic pancreas segmentation techniques have been actively developed in the research article for many years. Methods: Although the rapid growth of deep learning and proving satisfactory performance in many medical image processing and computer vision applications, the maximum Dice Similarity Coefficients (DSC) value of these techniques related to automatic pancreas segmentation is only around 85% due to complex structure of the pancreas and other factors. Contrary to previous techniques which are required significantly higher computational power and memory, this paper suggests a novel two-phase approach for high-accuracy automatic pancreas segmentation in computed tomography (CT) imaging. The proposed approach consists of two phases; (1) Pancreas Localization, where the rough pancreas position is detected on the 2D CT slice by adopting Mask R-CNN model, (2) Pancreas Segmentation, where the segmented pancreas region is produced by refining the candidate pancreas region with 3D U-Net on the 2D sub-CT slices generated in the first phase. Both qualitative and quantitative assessments of the proposed approach are performed on the NIH data set. Results: In order to evaluate the achievement of the recommended approach, a total of 16 different automatic pancreas segmentation techniques reported in the literature are compared by utilizing performance assessment procedures which are Dice Similarity Coefficient (DSC), Jaccard Index (JI), Precision (PRE), Recall (REC), Pixel Accuracy (ACC), Specificity (SPE), Receiver Operating Characteristics (ROC) and Area under ROC curve (AUC). The average values of DSC, JI, REC and ACC are computed as 86.15%, 75.93%, 86.27%, 86.27% and 99.95% respectively, which are the best values among well-established studies for automatic pancreas segmentation. Conclusion: It is demonstrated with qualitative and quantitative results that our suggested two-phase approach creates more favorable results than other existing approaches.

    • SDG 3 · Sağlık ve Kaliteli Yaşam
    İncele
  • Kitap BölümüYalnız künye2021
    Safer-Driving: Application of Deep Transfer Learning to Build Intelligent Transportation Systems

    Ünlü, Ramazan

    Deep learning methods are widely used in the field of transportation as well as in many areas. These days, deep learning methods produce very successful results thanks to big data, where data collection and storage are much easier than in past years. However, it can be quite difficult to collect data or find the right data in some cases. On the other hand, even if the desired data is collected enough, training deep neural networks to learn from these data may not be comfortable in terms of hardware needed for ordinary users. In these cases, transfer learning can be an exit point, which is very popular in deep understanding, especially when the amount of data is limited. In this study, we have aimed to give comprehensive guidance on using a transfer learning strategy to create a detection model to establish a safer transportation system. More specifically, we have developed a classification model to alleviate accidents due to deer on the road. We have implemented VGG-16 convolutional neural network architecture as a part of the structured transfer learning framework. Computational results show that high accuracy performance can be achieved by implementing VGG-16 with the respectively small size of a dataset.

    • SDG 9 · Sanayi, Yenilikçilik ve Altyapı
    • SDG 11 · Sürdürülebilir Şehirler ve Topluluklar
    İncele