Bilgisayar Mühendisliği Bölümü Koleksiyonu

Bu koleksiyon için kalıcı URI

https://hdl.handle.net/11468/10515

Listeleniyor 1 - 11 / 11

Classification of microscopic peripheral blood cell images using multibranch lightweight CNN-based model
(Springer Science and Business Media Deutschland GmbH, 2023) Fırat, Hüseyin
White blood cells (WBC), which are human peripheral blood cells, are the most significant part of the immune system that defends the body against microorganisms. Modifications in the morphological structure and number of subtypes of WBC play an major role in the diagnosis of serious diseases such as anemia and leukemia. Therefore, accurate WBC classification is clinically quite significant in the diagnosis of the disease. In last years, deep learning, especially CNN, has been used frequently in the field of medicine because of its strong self-learning capabilities and it can extract deeper features in images with stronger semantic information. In this study, a new CNN-based method is proposed for WBC classification. The proposed method (PM) is a hybrid method consisting of Inception module, pyramid pooling module (PPM) and depthwise squeeze-and-excitation block (DSEB). Inception module increases classification accuracy of CNNs by performing multiple parallel convolutions at different scales. PPM captures multi-scale contextual information from the input image by pooling features at multiple different scales. DSEB offers a structure where the network can selectively learn about informative features and remove useless ones. For the analysis of the classification results of the PM, experiments were carried out on three different datasets consisting of four classes (BCCD dataset), five classes (Raabin WBC dataset) and eight classes. As a result of the experimental studies, classification accuracy was obtained 99.96% in the BCCD dataset containing 4 classes, 99.22% in the Raabin WBC dataset containing 5 classes and 99.72% in the PBC dataset containing 8 classes. Compared with the state-of-the-art studies in the literature, the PM achieved the best accuracy in three datasets.
Retinal vessel segmentation via structure tensor coloring and anisotropy enhancement
(MDPI, 2017) Nergiz, Mehmet; Akın, Mehmet
Retinal vessel segmentation is one of the preliminary tasks for developing diagnosis software systems related to various retinal diseases. In this study, a fully automated vessel segmentation system is proposed. Firstly, the vessels are enhanced using a Frangi Filter. Afterwards, Structure Tensor is applied to the response of the Frangi Filter and a 4-D tensor field is obtained. After decomposing the Eigenvalues of the tensor field, the anisotropy between the principal Eigenvalues are enhanced exponentially. Furthermore, this 4-D tensor field is converted to the 3-D space which is composed of energy, anisotropy and orientation and then a Contrast Limited Adaptive Histogram Equalization algorithm is applied to the energy space. Later, the obtained energy space is multiplied by the enhanced mean surface curvature of itself and the modified 3-D space is converted back to the 4-D tensor field. Lastly, the vessel segmentation is performed by using Otsu algorithm and tensor coloring method which is inspired by the ellipsoid tensor visualization technique. Finally, some post-processing techniques are applied to the segmentation result. In this study, the proposed method achieved mean sensitivity of 0.8123, 0.8126, 0.7246 and mean specificity of 0.9342, 0.9442, 0.9453 as well as mean accuracy of 0.9183, 0.9442, 0.9236 for DRIVE, STARE and CHASE_DB1 datasets, respectively. The mean execution time of this study is 6.104, 6.4525 and 18.8370 s for the aforementioned three datasets respectively.
Modifiye edilmiş inception modülü kullanılarak insan periferik kan hücrelerinin çoklu sınıflandırılması
(Bandırma Onyedi Eylül Üniversitesi, 2023) Fırat, Hüseyin
Periferik kan hücrelerinin sınıflandırılması anemi ve lösemi gibi birçok kan hastalığının teşhisinde önemli rol oynamaktadır. Bu nedenle, doğru kan hücresi sınıflandırması hastalığın teşhisinde klinik olarak oldukça önemlidir. Son yıllarda, derin öğrenme, özellikle Evrişimsel sinir ağları, güçlü kendi kendine öğrenme yetenekleri sayesinde tıp alanında sıklıkla kullanılmaktadır. Bu çalışmada, kan hücre sınıflandırması için hesaplama maliyetini ve parametre sayısını azaltan derinlemesine ayrılabilir evrişim ile Inception modülünden oluşan yeni bir hibrit yöntem geliştirilmiştir. Bu yöntem, parametre sayısını ve hesaplama maliyetini azaltıp sınıflandırma doğruluğunu arttırmasıyla, standart evrişimsel sinir ağlarına göre bir avantaj sağlamaktadır. Geliştirilen yöntemin performansını test etmek için 8 sınıflı bir kan hücresi veri seti üzerinde yapılan deneysel çalışmalar sonucunda %98.89 doğruluk, %98.88 kesinlik, %98.85 duyarlılık, %98.86 F1-skoru elde edilmiştir. Literatürdeki çalışmalar ile karşılaştırıldığında yöntemimizin etkili olduğu görülmektedir.
Sıkma-uyarma artık ağı kullanılarak beyaz kan hücrelerinin sınıflandırılması
(Gazi Üniversitesi Bilişim Enstitüsü, 2023) Fırat, Hüseyin
Beyaz kan hücreleri, vücudun parazitler, bakteriler, virüsler gibi mikroorganizmalara karşı korunmasında etkin rol oynayan bağışıklık sisteminin önemli bir bileşenidir. Beyaz kan hücrelerinin yapısal özellikleri, alt türlerinin şekilleri ve sayıları insan sağlığı hakkında önemli bilgiler verebilmektedir. Hastalık teşhisinde doğru beyaz kan hücre tespiti klinik olarak oldukça önemlidir. Bu yüzden, doğru beyaz kan hücre sınıflandırma yöntemi kritik öneme sahiptir. Bu çalışmada, beyaz kan hücre sınıflandırması için Evrişimsel sinir ağı (ESA) tabanlı bir yöntem önerilmiştir. Önerilen yöntem sıkmauyarma ağı ile artık ağ mimarisinin birleşiminden oluşan hibrit bir yöntemdir. Derin ağ mimarilerinde katman sayısı arttıkça oluşabilecek problemler artık ağ ile çözülebilmektedir. Sıkma-uyarma (SU) bloğunun artık ağ ile birlikte kullanımı, toplam parametre sayısını minimum düzeyde arttırırken sınıflandırma doğruluğunu arttırmaktadır. Aynı zamanda, SU bloğunun artık ağ ile birleştirilmesi geleneksel artık ağların performansını da arttırmaktadır. Önerilen yöntemin performansını test etmek için Kaggle veritabanından alınan BCCD veriseti kullanılmıştır. Uygulamalar sonucunda ortalama %99,96 doğruluk, %99,92 kesinlik, duyarlılık ve F1-skoru elde edilmiştir. Bu sonuçlar, literatürden BCCD verisetini kullanan son yıllardaki çalışmalarda yer alan ESA yöntemlerinin elde ettiği sonuçlarla karşılaştırıldı ve önerilen yöntemin daha az eğitilebilir parametre ile daha iyi sonuçlar verdiği görülmüştür.
Federated learning-based colorectal cancer classification by convolutional neural networks and general visual representation learning
(John Wiley and Sons Inc., 2023) Nergiz, Mehmet
Colorectal cancer is the fourth fatal disease in the world, and the massive burden on the pathologists related to the classification of precancerous and cancerous colorectal lesions can be decreased by deep learning (DL) methods. However, the data privacy of the patients is a big challenge for being able to train deep learning models using big medical data. Federated Learning is a rising star in this era by providing the ability to train deep learning models on different sites without sacrificing data privacy. In this study, the Big Transfer model, which is a new General Visual Representation Learning method and six other classical DL methods are converted to the federated version. The effect of the federated learning is measured on all these models on four different data settings extracted from the MHIST and Chaoyang datasets. The proposed models are tested for single learning, centralized learning, and federated learning. The best AUC values of federated learning on Chaoyang are obtained by the Big Transfer and VGG models at 90.77% and 90.76%, respectively, whereas the best AUC value on MHIST is obtained by the Big Transfer model at 89.72%. The overall obtained results of models on all data settings show that the contribution of Federated Learning with respect to single learning is 4.71% and 11.68% for the “uniform” and “label-biased” data settings of Chaoyang, respectively, and 6.89% for the “difficulty level-biased” data setting of MHIST. Thus, it is experimentally shown that federated learning can be applied to the field of computational pathology for new institutional collaborations.
Collaborative colorectal cancer classification on highly class imbalanced data setting via federated neural style transfer based data augmentation
(International Information and Engineering Technology Association, 2022) Nergiz, Mehmet
The deep learning algorithms achieved promising results in the computational pathology in recent decade but the high data demand of the deep learning algorithms get stuck in the multi-institutional data collaborations. The federated learning is a novel concept, which proposes to train the models of the different sites collaboratively via an orchestrating server without leaking private data. However, the imbalanced data distributions are challenging for federated learning and result in performance decrease and destabilization. In this study, the federated version of the neural style transfer algorithm, which was offered by Gatys et al. is proposed as a data augmentation method on the highly class imbalanced configuration of Chaoyang colorectal cancer imaging dataset. The proposed method works by firstly selecting characteristic style images and then generating the gram style matrices on the local sites and then transferring them to the other imbalanced sites by not leaking any private data. The proposed method contributed the ACC, F1 Score and AUC results of pure FL by 22.07%, 42.51% and 9.65% using only 20 images for content and 5 images for style. Additionally, the experiments having different content and style numbers achieved the satisfactory and consisting results.
A hybrid DenseNet121-UNet model for brain tumor segmentation from MR Images
(Elsevier SCI LTD., 2022) Çınar, Necip; Özcan, Alper; Kaya, Mehmet
Several techniques are used to detect brain tumors in the medical research field; however, Magnetic Resonance Imaging (MRI) is still the most effective technique used by experts. Recently, researchers have proposed different MRI techniques to detect brain tumors with the possibility of uploading and visualizing the image. In the current decade, deep learning techniques have shown promising results in every research area, especially in bioinfor-matics and medical image analysis. This paper aims to segment brain tumors using deep learning methods of MR images. The UNet architecture, one of the deep learning networks, is used as a hybrid model with pre-trained DenseNet121 architecture for the segmentation process. During training and testing of the model, we focus on smaller sub-regions of tumors that comprise the complex structure. The proposed model is validated on BRATS 2019 publicly available brain tumor dataset that contains high-grade and low-grade glioma tumors. The experimental results indicate that our model performs better than other state-of-the-art methods presented in this particular area. Specifically, the best Dice Similarity Coefficient (DSC) are obtained by using the proposed approach to segment whole tumor (WT), core tumor (CT), and enhancing tumor (ET).
Facial landmark based region of interest localization for deep facial expression recognition
(Univ Osijek, 2022) Söylemez, Ömer Faruk; Ergen, Burhan
Automated facial expression recognition has gained much attention in the last years due to growing application areas such as computer animated agents, sociable robots and human computer interaction. The realization of a reliable facial expression recognition system through machine learning is still a challenging task particularly on databases with large number of images. Convolutional Neural Network (CNN) architectures have been proposed to deal with large numbers of training data for better accuracy. For CNNs, a task related best achieving architectural structure does not exist. In addition, the representation of the input image is equivalently important as the architectural structure and the training data. Therefore, this study focuses on the performances of various CNN architectures trained by different region of interests of the same input data. Experiments are performed on three distinct CNN architectures with three different crops of the same dataset. Results show that by appropriately localizing the facial region and selecting the correct CNN architecture it is possible to boost the recognition rate from 84% to 98% while decreasing the training time for proposed CNN architectures.
Detection of mitotic cells in breast cancer histopathological images using deep versus handcrafted features
(Springer, 2022) Sığırcı, İ. Onur; Albayrak, Abdülkadir; Bilgin, Gökhan
One of the most important processes in the diagnosis of breast cancer, which is the leading mortality rate in women, is the detection of the mitosis stage at the cellular level. In literature, many studies have been proposed on the computer-aided diagnosis (CAD) system for detecting mitotic cells in breast cancer histopathological images. In this study, comparative evaluation of conventional and deep learning based feature extraction methods for automatic detection of mitosis in histopathological images are focused. While various handcrafted features are extracted with textural/spatial, statistical and shape-based methods in conventional approach, the convolutional neural network structure proposed on the deep learning approach aims to create an architecture that extracts the features of small cellular structures such as mitotic cells. Mitosis detection/counting is an important process that helps us assess how aggressive or malignant the cancer's spread is. In the proposed study, approximately 180,000 non-mitotic and 748 mitotic cells are extracted for the evaluations. It is obvious that the classification stage cannot be performed properly due to the imbalanced numbers of mitotic and non-mitotic cells extracted from histopathological images. Hence, the random under-sampling boosting (RUSBoost) method is exploited to overcome this problem. The proposed framework is tested on mitosis detection in breast cancer histopathological images dataset provided from the International Conference on Pattern Recognition (ICPR) 2014 contest. In the results obtained with the deep learning approach, 79.42% recall, 96.78% precision and 86.97% F-measure values are achieved more successfully than handcrafted methods. A client/server-based framework has also been developed as a secondary decision support system for use by pathologists in hospitals. Thus, it is aimed that pathologists will be able to detect mitotic cells in various histopathological images more easily through necessary interfaces.
A whole-slide image grading benchmark and tissue classification for cervical cancer precursor lesions with inter-observer variability
(Springer Science and Business Media Deutschland GmbH, 2021) Albayrak, Abdulkadir; Akhan, Asli Ünlü; Çalık, Nurullah; Çapar, Abdulkerim; Bilgin, Gökhan; Töreyin, Behçet Uğur; Müezzinoğlu, Bahar; Türkmen, İlknur Çetinaslan; Durak Ata, Lütfiye
The cervical cancer developing from the precancerous lesions caused by the human papillomavirus (HPV) has been one of the preventable cancers with the help of periodic screening. Cervical intraepithelial neoplasia (CIN) and squamous intraepithelial lesion (SIL) are two types of grading conventions widely accepted by pathologists. On the other hand, inter-observer variability is an important issue for final diagnosis. In this paper, a whole-slide image grading benchmark for cervical cancer precursor lesions is created and the “Uterine Cervical Cancer Database” introduced in this article is the first publicly available cervical tissue microscopy image dataset. In addition, a morphological feature representing the angle between the basal membrane (BM) and the major axis of each nucleus in the tissue is proposed. The presence of papillae of the cervical epithelium and overlapping cell problems are also discussed. Besides that, the inter-observer variability is also evaluated by thorough comparisons among decisions of pathologists, as well as the final diagnosis.
Classification of analyzable metaphase images using transfer learning and fine tuning
(Springer, 2021) Albayrak, Abdulkadir
Chromosomes are bodies that contain human genetic information. Chromosomal disorders can cause structural and functional disorders in individuals. Detecting the metaphase stages of the cells accurately is a crucial step to detect possible defects in chromosomes. Thus, it is vital at this stage to identify the identical chromosome of each chromosome, to perform the pairing process, and to identify problems arising from this process. In this study, it was investigated whether the analyzable metaphase images can be analyzed by using the transfer learning and fine tuning approaches of deep learning models. The weights of VGG16 and InceptionV3 models trained with ImageNet data set were transferred to this problem and the classification process was carried out. True positive ratio values are 99%(+/- 0.9) and 99%(+/- 0.9) for VGG and Inception networks, respectively. The classification performances obtained depending on the changing training set ratios are presented comparatively in figures. F-measure, precision, and recall values obtained for the VGG and Inception networks were observed as 99%(+/- 1.0) and 99%(+/- 1.0), respectively. F-measure, precision, and recall values of VGG and Inceptionv3 networks are also presented with respect to the ratio of training size. The obtained results have compared with the state-of-the-art methods in the literature and supported with the tables and graphics. The training phase was also accelerated by using transfer learning and fine tuning methods. Transfer learning and fine tuning processes have almost similar performance as the models used in the literature and trained from scratch in metaphase Graphical Abstract The Flowchart of the proposed system for classifying metaphase candidates detection.

Güncel Gönderiler