Yazar "Ulukaya, Sezer" seçeneğine göre listele
Listeleniyor 1 - 20 / 25
Sayfa Başına Sonuç
Sıralama seçenekleri
Öğe Block Attention and Switchable Normalization Based Deep Learning Framework for Segmentation of Retinal Vessels(IEEE-Inst Electrical Electronics Engineers Inc, 2023) Deari, Sabri; Oksuz, Ilkay; Ulukaya, SezerThe presence of high blood sugar levels damages blood vessels and causes an eye condition called diabetic retinopathy. The ophthalmologist can detect this disease by looking at the variations in retinal blood vasculature. Manual segmentation of vessels requires highly skilled specialists, and not possible for many patients to be done quickly in their daily routine. For these reasons, it is of great importance to isolate retinal vessels precisely, quickly, and accurately. The difficulty distinguishing the retinal vessels from the background, and the small number of samples in the databases make this segmentation problem difficult. In this study, we propose a novel network called Block Feature Map Distorted Switchable Normalization U-net with Global Context Informative Convolutional Block Attention Module (BFMD SN U-net with GCI- CBAM). We improve the traditional Fully Convolutional Segmentation Networks in multiple aspects with the proposed model as follows; The model converges in earlier epochs, adapts more flexibly to different data, is more robust against overfitting, and gets better feature refinement at different dilation rates to cope with different sizes of retinal vessels. We evaluate the proposed network on two reference retinal datasets, DRIVE and CHASE DB1, and achieve state-of-the-art performances with 97.00 % accuracy and 98.71 % AUC in DRIVE and 97.62 % accuracy and 99.11 % AUC on CHASE DB1 databases. Additionally, the convergence step of the model is reduced and it has fewer parameters than the baseline U-net. In summary, the proposed model surpasses the U-net -based approaches used for retinal vessel separation in the literature.Öğe Classification of Parkinson's Disease Using Dynamic Time Warping(IEEE, 2019) Kurt, Ilke; Ulukaya, Sezer; Erdem, OguzhanDeteriorations in handwriting or in basic shape sketching are one of the most referenced symptoms for early diagnosis of Parkinson's disease (PD). For this reason, the design of a fair, trustworthy and efficacious Computer-aided Diagnosis (CAD) model has supportive importance for the early diagnosis of PD. In this study we investigate the effectiveness of Dynamic Time Warping (DTW) algorithm, which is applied to Archimedean spiral drawings of patients with PD and healthy controls (HC), on PD and healthy subject classification problem. Leave-one-subject-out (LOSO) cross validation scheme is used while training and testing in support vector machine (SVM) and k-nearest neighbors (k-NN) classifiers with various parameters. The accuracy results of %94.44 (%95.83) and %97.52 (%94.44) are achieved by k-NN and SVM classifiers respectively for static (dynamic) spiral test.Öğe A Comparison of Facial Landmark Detection Methods(IEEE, 2018) Sandikci, Esra Nur; Eroglu Erdem, Cigdem; Ulukaya, SezerFace analysis is a rapidly developing research area and facial landmark detection is one of the pre-processing steps. In recent years, many algorithms and comprehensive survey/challenge papers have been published on facial landmark detection. In this work, we analysed six survey/challenge papers and observed that among open source systems deep learning (TCDCN, DCR) and regression based (CFSS) methods show superior performance.Öğe Consensus and stacking based fusion and survey of facial feature point detectors(Springer Heidelberg, 2022) Ulukaya, Sezer; Sandikci, Esra Nur; Erdem, Cigdem ErogluFacial landmark detection is a crucial pre-processing step for many applications including face tracking, face recognition and facial affect recognition. Hence, we first aim to investigate and experimentally compare the most successful open source facial feature point detection algorithms published in the last decade. We first present an overview of surveys on facial feature detection algorithms to provide insight into the challenges and innovations. We also propose a consensus-based selection and stacked regression based fusion of facial landmark methods to combine their results in order to achieve superior accuracy. Five open-source algorithms in the literature are objectively compared using the same test data and regression based models have been shown to be more successful. According to the extensive experimental results, the proposed consensus and stacking based fusion method gives the lowest facial landmark detection error as compared to the five most successful algorithms in the literature. Consensus and stacking based fusion of an ensemble of methods boosts the performance of facial landmark detection. The proposed fusion method can also be applied future methods as they emerge.Öğe Deep Learning based Classification of Human Nail Diseases using Color Nail Images(IEEE, 2022) Yamac, Senar Ali; Kuyucuoglu, Orhun; Koseoglu, Seyma Begum; Ulukaya, SezerWhen the disorders that occur in the fingernails and toenails are not noticed early, they can turn into diseases that affect human life. These diseases in our hands and feet, which are mostly used organs in our daily work, also negatively affect the quality of life. In this study, it is aimed to detect 5 different nail diseases using deep learning architectures. Within the scope of the study, the performance of the 6 most recent deep learning architectures was compared with each other. Although the number of pictures in the open-access database used in the study is low, the obtained results seem to be successful.Öğe DETECTION OF NAIL DISEASES USING ENSEMBLE MODEL BASED ON MAJORITY VOTING(2023) Yamaç, Senar Ali; Kuyucuoğlu, Orhun; Köseoğlu, Şeyma Begüm; Ulukaya, SezerNail diseases are disorders that can have serious effects on human quality of life. With the developing computational methods and technology, anomalies on the nail may be detected quickly and in a non-invasive way. This study proposes a model that provides better performance by combining the results of different deep learning networks with the ensemble learning method. The performance of 7 different deep learning architectures was examined using a database containing 17 disease classes. The proposed method achieved 75 % accuracy, resulting in significant increases in precision and recall metrics compared to individual deep-learning architectures. Thanks to a mobile application that will be developed, the proposed model for large-scale screening may be used as an assistive decision support system for medical professionals. When the results are observed, we predict that early detection of nail diseases (in a remote way) on the hand, which is one of our most used limbs, can reduce hospital visits and costs. In addition, the proposed method can be integrated into dermatoscopy devices used for skin diseases and mole analysis.Öğe Detection of Parkinson's disease with keystroke data(Taylor & Francis Ltd, 2023) Demir, Bahar; Ulukaya, Sezer; Erdemb, OguzhanParkinson's disease (PD) is one of the most widespread neurological disorders associated with nerve damage without definitive treatment. Impairments, such as trembling and slowing down in hand movements are among the first symptoms. For this purpose, in this study, machine learning (ML)-based models were developed by using keyboard keystroke dynamics. According to patients' drug use status, disease severity, and gender, we created 14 different sub-datasets and extracted 378 features using raw keystroke data. We developed alternative models with Support Vector Machines (SVM), k-Nearest Neighbors (kNN), and Random Forest (RF) algorithms. We further used Minimum Redundancy Maximum Relevance (mRmR), RELIEF, sequential forward selection (SFS), and RF feature selection methods to investigate prominent features in distinguishing PD. We developed ML models that jointly use the most popular features of selection algorithms (feature ensemble [FE]) and an ensemble classifier by combining multiple ML algorithms utilizing majority vote (model ensemble [ME]). With 14 different sets, FE and ME models provided accuracy (Acc.) in the range of 91.73 - 100% and 81.08 - 100%, respectively.Öğe Feature Extraction Using Time-Frequency Analysis for Monophonic-Polyphonic Wheeze Discrimination(IEEE, 2015) Ulukaya, Sezer; Sen, Ipek; Kahya, Yasemin P.The aim of this study is monophonic-polyphonic wheeze episode discrimination rather than the conventional wheeze (versus non-wheeze) episode detection. We used two different methods for feature extraction to discriminate monophonic and polyphonic wheeze episodes. One of the methods is based on frequency analysis and the other is based on time analysis. Frequency analysis based method uses ratios of quartile frequencies to exploit the difference in the power spectrum. Time analysis based method uses mean crossing irregularity to exploit the difference in periodicity in the time domain. Both methods are applied on the data before and after an image processing based preprocessing step. Calculated features are used in classification both individually and in combinations. Support vector machine, k-nearest neighbor and Naive Bayesian classifiers are adopted in leave-one-out scheme. A total of 121 monophonic and 110 polyphonic wheeze episodes are used in the experiments, where the best classification performances are 71.45% for time domain based features, 68.43% for frequency domain based features, and 75.78% for a combination of selected best features.Öğe Gait analyses in multiple sclerosis; and time series(Iop Publishing Ltd, 2020) Keklicek, Hilal; Ulukaya, Sezer[Abstract Not Available]Öğe A Hybrid Fusion Method Combining Spatial Image Filtering with Parallel Channel Network for Retinal Vessel Segmentation(Springer Heidelberg, 2023) Yakut, Cem; Oksuz, Ilkay; Ulukaya, SezerRetinography is a frequently used imaging method that aids in the clinical diagnosis of eye disorders. Low contrast and being exposed to noise are the primary factors in degraded retinal fundus images. These factors make it challenging for medical experts to diagnose and classify diseases in retinal images. This manuscript proposes a hybrid fusion approach for vascular tree segmentation in color fundus images. We propose to use a fusion model that combines supervised deep convolutional neural networks with unsupervised approaches. The training fundus images were preprocessed in an unsupervised way to increase the success of the deep U-Net architecture and fed into the network as parallel channels. Preprocessing steps include the following stages: grayscale conversion, median filtering, CLAHE, mathematical morphology operations, Coye filtering, connected component analysis, and data augmentation. The proposed approach was tested on publicly accessible DRIVE and HRF datasets. Sensitivity, specificity, accuracy, and F1-score measures are compared on high and low-resolution datasets. In summary, results reveal that the performance of the parallel channel-based deep approach is higher than the baseline deep model and achieved state-of-the-art results in the literature, especially on the HRF dataset. Besides, the fusion of the predictions of only the unsupervised image processing-based models achieved the best accuracy among unsupervised works in the literature on the DRIVE dataset. Moreover, the proposed unsupervised preprocessing-based approach does not add a significant computational burden on the deep learning model training. Additionally, the proposed hybrid method has noticeably increased the sensitivity rate on both datasets.Öğe A Hybrid Multistage Model Based on YOLO and Modified Inception Network for Rice Leaf Disease Analysis(Springer Heidelberg, 2024) Deari, Sabri; Ulukaya, SezerLack or excess of water, moisture, and nutrients may cause diseases in various growing stages of rice. Unlike related studies, this work aims to detect each disease's symptom separately, rather than just classifying images by a classifier or showing the whole diseased leaf in a single bounding box. In this way, we consider all disease regions and make it possible to observe better the disease progression by considering detected boxes. Our motivation for this hybrid study stems from the fact that more than one disease symptom may occur on a leaf and the detection of symptoms at an early stage can positively affect the harvest yield. The main aim of this study is to classify rice leaf disease images accurately, reduce false detections, and validate the predictions of the classification network utilizing an object detection network. Therefore, we identify two stages for this work. In the first part, the task of classification, and in the second part, the task of determining the location of the disease symptoms is conducted. We use data augmentation and disout techniques to prevent overfitting in the classification process and to improve performance by modifying the classification network. Finally, we discuss how classification robustness can be tested and false predictions can be eliminated using the classification network Inception v3 and the detection network YOLOv5x jointly. As a result of the proposed hybrid model, state-of-the-art results are achieved with 96.67 % accuracy and 98.24 % F1 score on the publicly available rice leaf disease dataset.Öğe Importance of Data Augmentation and Transfer Learning on Retinal Vessel Segmentation(IEEE, 2021) Deari, Sabri; Oksuz, Ilkay; Ulukaya, SezerAutomatic segmentation of retinal fundus images for extracting blood vessels is an essential task in the diagnostic classification of hypertension, glaucoma, and diabetic retinopathy, which are the leading causes of blindness. In this paper, we employed a transfer learning strategy for improved retinal vessel extraction. Firstly, we trained the U-NET model on CHASE DB1 and DRIVE databases. By using data augmentations on datasets we enable the U-NET model to learn retinal vessel features better. We examined the data augmentation types, namely, pixel-level transformations and affine transformations. Secondly, we utilized the transfer learning approach on two datasets and achieved comparable results with the state-of-the-art studies on retinal vessel segmentation task. Also, we employed combination of affine and pixel-level transformations to further boost segmentation performance.Öğe Individuals with a COVID-19 history exhibit asymmetric gait patterns despite full recovery(Elsevier Sci Ltd, 2022) Keklicek, Hilal; Selcuk, Halit; Kurt, Ilke; Ulukaya, Sezer; Ozturk, GulnurCOVID-19 is a multisystem infectious disease affecting the body systems. Its neurologic complications include -but are not limited to headache, loss of smell, encephalitis, and cerebrovascular accidents. Even though gait analysis is an objective measure of the neuro-motor system and may provide significant information about the pathophysiology of specific diseases, no studies have investigated the gait characteristics in adults after full recovery from COVID-19. This was a cross-sectional, controlled study that included 12 individuals (mean age, 23.0 +/- 4.1 years) with mild-to-moderate COVID-19 history (COVD) and 20 sedentary controls (CONT; mean age, 24.0 +/- 3.6 years). Gait was evaluated using inertial sensors on a motorized treadmill. Spatial-temporal gait parameters and gait symmetry were calculated by using at least 512 consecutive steps for each participant. The effect-size analyses were utilized to interpret the impact of the results. Spatial-temporal gait characteristics were comparable between the two groups. The COVD group showed more asymmetrical gait patterns than the CONT group in the double support duration symmetry (p = 0.042), single support duration symmetry (p = 0.006), loading response duration symmetry (p = 0.042), and pre-swing duration symmetry (p = 0.018). The effect size analyses of the differences showed large effects (d = 0.68-0.831). Individuals with a history of mild-to-moderate COVID-19 showed more asymmetrical gait patterns than individuals without a disease history. Regardless of its severity, the multifaceted long-term effects of COVID-19 need to be examined and the scope of clinical follow-up should be detailed.Öğe A Lung Sound Classification System based on the Rational Dilation Wavelet Transform(IEEE, 2016) Ulukaya, Sezer; Serbes, Gorkem; Sen, Ipek; Kahya, Yasemin P.I n this work, a wavelet based classification system that aims to discriminate crackle, normal and wheeze lung sounds is presented. While the previous works related with this problem use constant low Q-factor wavelets, which have limited frequency resolution and can not cope with oscillatory signals, in the proposed system, the Rational Dilation Wavelet Transform, whose Q-factors can be tuned, is employed. Proposed system yields an accuracy of 95 % for crackle, 97 % for wheeze, 93.50 % for normal and 95.17 % for total sound signal types using energy feature subset and proposed approach is superior to conventional low Q-factor wavelet analysis.Öğe MSCCov19Net: multi-branch deep learning model for COVID-19 detection from cough sounds(Springer Heidelberg, 2023) Ulukaya, Sezer; Sarica, Ahmet Alp; Erdem, Oguzhan; Karaali, AliCoronavirus has an impact on millions of lives and has been added to the important pandemics that continue to affect with its variants. Since it is transmitted through the respiratory tract, it has had significant effects on public health and social relations. Isolating people who are COVID positive can minimize the transmission, therefore several exams are proposed to detect the virus such as reverse transcription-polymerase chain reaction (RT-PCR), chest X-Ray, and computed tomography (CT). However, these methods suffer from either a low detection rate or high radiation dosage, along with being expensive. In this study, deep neural network-based model capable of detecting coronavirus from only coughing sound, which is fast, remotely operable and has no harmful side effects, has been proposed. The proposed multi-branch model takes Mel Frequency Cepstral Coefficients (MFCC), Spectrogram, and Chromagram as inputs and is abbreviated as MSCCov19Net. The system is trained on publicly available crowdsourced datasets, and tested on two unseen (used only for testing) clinical and non-clinical datasets. Experimental outcomes represent that the proposed system outperforms the 6 popular deep learning architectures on four datasets by representing a better generalization ability. The proposed system has reached an accuracy of 61.5 % in Virufy and 90.4 % in NoCoCoDa for unseen test datasets.Öğe Musical Feature Based Classification of Parkinson's Disease Using Dysphonic Speech(IEEE, 2018) Kurt, Ilke; Ulukaya, Sezer; Erdem, OguzhanSpeech and voice disorders are one of the most significant biomarkers in early diagnosis of Parkinson's disease (PD). The development of an objective, reliable and effective prediction model is crucial for the early detection of PD by experts. The aim of this study is to investigate the effectiveness of musical features of voice recordings on PD and healthy subject discrimination issue. Extracted number of 41 musical features from the voice recordings of 28 PD and 62 healthy controls are used in the context of music information retrieval. These features are employed in the classification models either as a single large set or partitioned into smaller feature groups. Leave-one-subject-out (LOSO), leave-one-out (LOO) and 10-fold cross validation schemes are used while training and testing in support vector machine (SVM) and k-nearest neighbors (k-NN) classifiers by providing statistical measures. The effect of low, normal and high tone voice recordings is also studied separately, and the results show that using low-tone voice recordings may not be useful for discrimination of dysphonic voice. Despite using least number of features of all related schemes which use raw voice recordings, our proposed musical features with LOSO cross validation technique perform better accuracy results than the existing studies.Öğe A Novel Method for Determination of Wheeze Type(IEEE, 2015) Ulukaya, Sezer; Sen, Ipek; Kahya, Yasemin P.Among respiratory disorders, obstructive diseases such as asthma and chronic obstructive pulmonary disease (COPD) constitute an important group. To our knowledge, there does not exist a study in the literature that quantifies the relationship between the type of wheeze and the type or severity of the disease. This study, aims at classifying wheeze type rather than classical normal-wheeze sound classification studies in the literature. In this study, we propose a method based on Multiple Signal Classification (MUSIC) algorithm to differentiate between monophonic and polyphonic wheezes, without a need for pre-training the algorithm. The algorithm determines the true labels of monophonic and polyphonic wheezes with 100% and 78% accuracy, respectively. Since there does not exist a method in the literature that has been proposed specifically for this problem, only the results of the most relevant few studies have been presented. Since the proposed system can directly estimate the frequency, we consider the method proposed here would be a useful quantification method for further studies in medical literature, on finding correlations between wheezes and disorders.Öğe An open access database for the evaluation of respiratory sound classification algorithms(Iop Publishing Ltd, 2019) Rocha, Bruno M.; Filos, Dimitris; Mendes, Luis; Serbes, Gorkem; Ulukaya, Sezer; Kahya, Yasemin P.; Jakovljevic, NiksaObjective: Over the last few decades, there has been significant interest in the automatic analysis of respiratory sounds. However, currently there are no publicly available large databases with which new algorithms can be evaluated and compared. Further developments in the field are dependent on the creation of such databases. Approach: This paper describes a public respiratory sound database, which was compiled for an international competition, the first scientific challenge of the IFMBE' s International Conference on Biomedical and Health Informatics. The database includes 920 recordings acquired from 126 participants and two sets of annotations. One set contains 6898 annotated respiratory cycles, some including crackles, wheezes, or a combination of both, and some with no adventitious respiratory sounds. In the other set, precise locations of 10 775 events of crackles and wheezes were annotated. Main results: The best system that participated in the challenge achieved an average score of 52.5% with the respiratory cycle annotations and an average score of 91.2% with the event annotations. Significance: The creation and public release of this database will be useful to the research community and could bring attention to the respiratory sound classification problem.Öğe Overcomplete discrete wavelet transform based respiratory sound discrimination with feature and decision level fusion(Elsevier Sci Ltd, 2017) Ulukaya, Sezer; Serbes, Gorkem; Kahya, Yasemin P.Background and objective: Crackle, wheeze and normal lung sound discrimination is vital in diagnosing pulmonary diseases. Previous works suffer from limited frequency resolution and lack of the ability to deal with oscillatory signals (wheezes). The main objective of this study is to propose a novel wavelet based lung sound classification system that is capable of adaptively representing crackle, wheeze and normal lung sound signal time-frequency properties. Methods: A method which is based on rational dilation wavelet transform is proposed to classify lung sounds into three main categories, namely, normal, wheeze and crackle. Six different feature extraction methods were used with five different classifiers all of which were compared with the proposed method on 600, lung sound episodes in a cross validation scheme. Six statistical subset features were extracted from raw features and fed into classifiers. After comparative evaluation of the proposed method, an ensemble learning scheme was built to increase the performance of the proposed method. Results: It has been shown that performance of the proposed method was superior to previous methods in terms of accuracy. Moreover, its computational time was far less than its nearest competitor (S transform). It has also been shown that the proposed method was able to cope with oscillatory type signals as well as transient sounds performing 95.17% average accuracy for energy subset and 97.38% ensemble average accuracy showing a promising time-frequency tool for biological signals. Conclusions: The proposed method has shown better performance even using only one subset of extracted features. It provides better time-frequency resolution for all types of signals of interest and is less redundant than continuous wavelet transform and significantly faster than its nearest competitor. (C) 2017 Elsevier Ltd. All rights reserved.Öğe Periodontal bone loss detection based on hybrid deep learning and machine learning models with a user-friendly application(Elsevier Sci Ltd, 2022) Sunnetci, Kubilay Muhammed; Ulukaya, Sezer; Alkan, AhmetAs artificial intelligence in medical imaging is used to diagnose many diseases, it can also be employed to diagnose whether a person has periodontal bone loss or not. Accurate and early diagnosis performs a vital task in the treatment of the patient's dental disorder. Therefore, such medical images are known to be an important clinical adjunct. In this manuscript, whether the patient has periodontal bone loss or non-periodontal bone loss is diagnosed employing hybrid artificial intelligence-based systems. Herein, after tagging a total of 1432 images by an expert, we extract 1000 deep image features for each image using AlexNet and SqueezeNet deep learning architectures. On the other hand, we classify these images directly without extracting the image features using the EfficientNetB5 deep learning architecture. First, we categorize AlexNet-based deep image features using the Coarse Tree, Weighted K-Nearest Neighbor (KNN), Gaussian Naive Bayes, RUSBoosted Trees Ensemble, and Linear Support Vector Machine (SVM) classifiers. Afterward, we classify SqueezeNet-based deep image features using Medium Tree, Gaussian Naive Bayes, Boosted Trees Ensemble, Coarse KNN, and Medium Gaussian SVM classifiers. With the help of the ten classifiers employed in this study, we also design a user-friendly Graphical User Interface (GUI) application. Thanks to this application, we aim to reduce the workload of experts, save time and help to diagnose dental disorders early. The results show that the best classifiers for AlexNet-based, SqueezeNet-based, and Direct-Convolutional Neural Network (CNN) are Linear SVM, Medium Gaussian SVM, and EfficientNetB5, respectively. Among these classifiers, the best classifier is Linear SVM, and its accuracy, error, sensitivity, specificity, precision, and F1 score values are 81.49%, 18.51%, 84.57%, 79.14%, 75.68%, and 79.88%, respectively.
