Browsing by Author "Zan, Hasan"

Now showing 1 - 12 of 12

Abnormal Heart Sound Detection Using Ensemble Classifiers
(IEEE, 2018) Zan, Hasan; Yıldız, Abdulnasır; Zan, Hasan
Phonocardiogram is used for ambulatory diagnostic to assess health status of heart and detect cardiovascular disease. The goal of this study is to develop automatic classification method of PCG recordings collected from different databases and recorded in a different way. For this purpose, after various time and frequency domain features are extracted from PCG recordings obtained from two databases, recordings are subjected to pre-classification in order determine which database they are obtained from. Before final classification, various time, frequency and time-frequency domain features of classified recordings are extracted. These features are fed into four different classification ensembles trained with training dataset. With final decision rule, proposed algorithm achieved an accuracy of 98.9%, a sensitivity of 93.75% and a specify of 99.5%.
Classification and analysis of epileptic EEG recordings using convolutional neural network and class activation mapping
(2021) Yildiz, Abdulnasir; Zan, Hasan; Said, Sherif; Zan, Hasan
Electrical bio-signals have the potential to be used in different applications due to their hidden nature and their ability to facilitate liveness detection. This paper investigates the feasibility of using the Convolutional Neural Network (CNN) to classify and analyze electroencephalogram (EEG) data with their time-frequency representations and class activation mapping (CAM) to detect epilepsy disease. Several types of pre-trained CNNs are employed for a multi-class classification task (AlexNet, GoogLeNet, ResNet-18, and ResNet-50) and their results are compared. Also, a novel convolutional neural network architecture comprised of two horizontally concatenated GoogLeNets is proposed with two inputs scalograms and spectrogram of the eplictic EEG signal. Four segment lengths (4097, 2048, 1024, and 512 sampling points) with three time-frequency representations (short-time Fourier, Wavelet, and Hilbert-Huang transform) are statistically evaluated. The dataset used in this research is collected at the University of Bonn. The dataset is reorganized as normal, interictal, and ictal. The maximum achieved accuracies for 4097, 2048, 1024, and 512 sampling points are 100 %, 100 %, 100 %, and 99.5 % respectively. The CAM method is used to analyze discriminative regions of time-frequency representations of EEG segments and networks' decisions. This method showed CNN models used different time and frequency regions of input images for each class with correct and incorrect predictions.
Classification and analysis of epileptic EEG recordings using convolutional neural network and class activation mapping
(Biomedical Signal Processing and Control, 2021) Zan, Hasan; Yıldız, Abdulnasir; Said, Sherif; Zan, Hasan
Electrical bio-signals have the potential to be used in different applications due to their hidden nature and their ability to facilitate liveness detection. This paper investigates the feasibility of using the Convolutional Neural Network (CNN) to classify and analyze electroencephalogram (EEG) data with their time-frequency representations and class activation mapping (CAM) to detect epilepsy disease. Several types of pre-trained CNNs are employed for a multi-class classification task (AlexNet, GoogLeNet, ResNet-18, and ResNet-50) and their results are compared. Also, a novel convolutional neural network architecture comprised of two horizontally concatenated GoogLeNets is proposed with two inputs scalograms and spectrogram of the eplictic EEG signal. Four segment lengths (4097, 2048, 1024, and 512 sampling points) with three time-frequency representations (short-time Fourier, Wavelet, and Hilbert-Huang transform) are statistically evaluated. The dataset used in this research is collected at the University of Bonn. The dataset is reorganized as normal, interictal, and ictal. The maximum achieved accuracies for 4097, 2048, 1024, and 512 sampling points are 100 %, 100 %, 100 %, and 99.5 % respectively. The CAM method is used to analyze discriminative regions of time-frequency representations of EEG segments and networks' decisions. This method showed CNN models used different time and frequency regions of input images for each class with correct and incorrect predictions.
Classification of Eeg Signals Using Hilbert-Huang Transform-Based Deep Neural Networks
(IEEE, 2019) Zan, Hasan; Yildiz, Abdulnasir; Ozerdem, Mehmet Sirac; Zan, Hasan
Epilepsy is one of the most common neurologic disease. Electroencephalogram (EEG) contains physiologic and pathological information related human nervous system. EEG signals used in this study are obtained from Bonn University, Department of Epileptology EEG database. Original database has five subsets (A, B, C, D, E). Data have been reorganized into three groups which are healthy (AB), interictal (CD) and ictal EEG signals. Furthermore, in order to examine effect of signal length on classification performance, three different lengths are used. Hilbert-Huang transform is applied to the signals and they are represented as image files. Then, generated images are fed into deep neural networks with five different structures for classification. Accuracy is calculated for all cases to asses performance of proposed method. it is clear that successful results could be obtained using Hilbert-Huang transform along with deep learning networks.
Engine Fault Detection by Sound Analysis and Machine Learning
(Mdpi, 2024) Akbalik, Ferit; Yildiz, Abdulnasir; Ertugrul, Omer Faruk; Zan, Hasan; Zan, Hasan
Traditional vehicle fault diagnosis methods rely heavily on the expertise of mechanics or diagnostic tools available at service centers, which can be costly, time-consuming, and may not always provide accurate results. This study presents a comprehensive vehicle fault diagnosis framework, which utilized Mel-Frequency Cepstral Coefficients (MFCCs), Discrete Wavelet Transform (DWT)-based features, and the Extreme Learning Machine (ELM) classifier. To address the limitations of previous works, the proposed framework leverages a large, diverse dataset encompassing various vehicle models and real-world operating conditions. Significantly improved robustness and generalizability of the fault diagnosis system were achieved. The results of the experiments demonstrate the superiority of the MFCC-based features combined with the ELM classifier, achieving the highest performance metrics in terms of accuracy, precision, recall, F1-score, macro F1-score, and weighted F1-score, which are 92.17%, 92.24%, 92.22%, 92.10%, and 92.06%, respectively. Slightly lower performance was obtained while employing the DWT-based features compared to employing MFCC-based features. Additionally, frequency analysis was conducted to identify specific frequency bins, which are the most indicative of different fault types in providing valuable guidance for future diagnostic efforts. Overall, the proposed framework provides a reliable and practical solution for accurate vehicle fault detection, paving the way for future advancements in automotive diagnostics.
Enhancing Vehicle Fault Diagnosis Through Multi-View Sound Analysis: Integrating Scalograms and Spectrograms in a Deep Learning Framework
(Springer London Ltd, 2025) Akbalik, Ferit; Yildiz, Abdulnasir; Ertugrul, Omer Faruk; Zan, Hasan; Zan, Hasan
This study presents a comprehensive framework for vehicle fault diagnosis using engine sound signals, leveraging deep learning models and a multi-view approach. Traditional methods for vehicle fault diagnosis often rely on the expertise of mechanics or diagnostic tools, which can be costly, time-consuming, and may not always provide accurate results. To address these limitations, we propose CarFaultNet, a multi-view model that processes both scalograms and spectrograms simultaneously to capture complementary information from these time-frequency representations. Our approach incorporates transfer learning with pretrained convolutional neural networks, including AlexNet, GoogLeNet, ShuffleNet, SqueezeNet, and MobileNet v2, as well as CarFaultNet, which combines two MobileNet networks. The results demonstrate that CarFaultNet outperforms traditional machine learning methods and single-view deep learning models, achieving a precision of 95.32%, recall of 94.83%, F1-score of 94.99%, and accuracy of 95.00%. Class activation mapping visualizations provide valuable insights into the model's decision-making process, highlighting the regions of the input images that are most influential for the classification of different vehicle faults. By leveraging a large, diverse dataset encompassing various vehicle models and real-world operating conditions, our approach addresses the drawbacks of previous studies and demonstrates the potential of deep learning for practical and effective vehicle fault diagnosis.
Local Pattern Transformation-Based convolutional neural network for sleep stage scoring
(2023) Zan, Hasan; Yıldız, Abdulnasır; Zan, Hasan
Sleep stage scoring is essential for the diagnosis and treatment of sleep disorders. However, manual sleep scoring is a tedious, time-consuming, and subjective task. Therefore, this paper proposes a novel framework based on local pattern transformation (LPT) methods and convolutional neural networks for automatic sleep stage scoring. Unlike in previous works in other fields, these methods were not employed for manual feature extraction, which requires expert knowledge and the pipeline behind it might bias results. The transformed signals were directly fed into a CNN model (called EpochNet) that can accept multiple successive epochs. The model learns features from multiple input epochs and considers inter-epoch context during classification. To evaluate and validate the effectiveness of the proposed approach, we conducted several experiments on the Sleep-EDF dataset. Four LPT methods, including One-dimensional Local Binary Pattern (1D-LBP), Local Neighbor Descriptive Pattern (LNDP), Local Gradient Pattern (LGP), and Local Neighbor Gradient Pattern (LNGP), and different polysomnography (PSG) signals were analyzed as sequence length (number of input epochs) increased from one to five. 1D-LBP and LNDP achieved similar performances, outperforming other LPT methods that are less sensitive to local variations. The best performance was achieved when an input sequence containing five epochs of PSG signals transformed by 1D-LBP was employed. The best accuracy, F1 score, and Kohen's kappa coefficient were 0.848, 0.782, and 0.790, respectively. The results showed that our approach can achieve comparable performance to other state-of-the-art methods while occupying fewer computing resources because of the compact size of EpochNet.
Local Pattern Transformation-Based convolutional neural network for sleep stage scoring
(ScienceDirect, 2023) Zan, Hasan; Yildiz, Abdulnasır; Zan, Hasan
Sleep stage scoring is essential for the diagnosis and treatment of sleep disorders. However, manual sleep scoring is a tedious, time-consuming, and subjective task. Therefore, this paper proposes a novel framework based on local pattern transformation (LPT) methods and convolutional neural networks for automatic sleep stage scoring. Unlike in previous works in other fields, these methods were not employed for manual feature extraction, which requires expert knowledge and the pipeline behind it might bias results. The transformed signals were directly fed into a CNN model (called EpochNet) that can accept multiple successive epochs. The model learns features from multiple input epochs and considers inter-epoch context during classification. To evaluate and validate the effectiveness of the proposed approach, we conducted several experiments on the Sleep-EDF dataset. Four LPT methods, including One-dimensional Local Binary Pattern (1D-LBP), Local Neighbor Descriptive Pattern (LNDP), Local Gradient Pattern (LGP), and Local Neighbor Gradient Pattern (LNGP), and different polysomnography (PSG) signals were analyzed as sequence length (number of input epochs) increased from one to five. 1D-LBP and LNDP achieved similar performances, outperforming other LPT methods that are less sensitive to local variations. The best performance was achieved when an input sequence containing five epochs of PSG signals transformed by 1D-LBP was employed. The best accuracy, F1 score, and Kohen’s kappa coefficient were 0.848, 0.782, and 0.790, respectively. The results showed that our approach can achieve comparable performance to other state-ofthe-art methods while occupying fewer computing resources because of the compact size of EpochNet.
Multi-task learning for arousal and sleep stage detection using fully convolutional networks
(IOP Publishing, 2023) Zan, Hasan; Yıldız, Abdulnasir; Zan, Hasan
Objective: Sleep is a critical physiological process that plays a vital role in maintaining physical and mental health. Accurate detection of arousals and sleep stages is essential for the diagnosis of sleep disorders, as frequent and excessive occurrences of arousals disrupt sleep stage patterns and lead to poor sleep quality, negatively impacting physical and mental health. Polysomnography is a traditional method for arousal and sleep stage detection that is time-consuming and prone to high variability among experts. Approach: In this paper, we propose a novel multi-task learning approach for arousal and sleep stage detection using fully convolutional neural networks. Our model, FullSleepNet, accepts a full-night single-channel EEG signal as input and produces segmentation masks for arousal and sleep stage labels. FullSleepNet comprises four modules: a convolutional module to extract local features, a recurrent module to capture long-range dependencies, an attention mechanism to focus on relevant parts of the input, and a segmentation module to output final predictions. Main results: By unifying the two interrelated tasks as segmentation problems and employing a multi-task learning approach, FullSleepNet achieves state-of-the-art performance for arousal detection with an area under the precision-recall curve of 0.70 on Sleep Heart Health Study and Multi-Ethnic Study of Atherosclerosis datasets. For sleep stage classification, FullSleepNet obtains comparable performance on both datasets, achieving an accuracy of 0.88 and an F1-score of 0.80 on the former and an accuracy of 0.83 and an F1-score of 0.76 on the latter. Significance: Our results demonstrate that FullSleepNet offers improved practicality, efficiency, and accuracy for the detection of arousal and classification of sleep stages using raw EEG signals as input.
Multi-task learning for arousal and sleep stage detection using fully convolutional networks
(2023) Zan, Hasan; Yıldız, Abdulnasır; Zan, Hasan
Objective. Sleep is a critical physiological process that plays a vital role in maintaining physical and mental health. Accurate detection of arousals and sleep stages is essential for the diagnosis of sleep disorders, as frequent and excessive occurrences of arousals disrupt sleep stage patterns and lead to poor sleep quality, negatively impacting physical and mental health. Polysomnography is a traditional method for arousal and sleep stage detection that is time-consuming and prone to high variability among experts. Approach. In this paper, we propose a novel multi-task learning approach for arousal and sleep stage detection using fully convolutional neural networks. Our model, FullSleepNet, accepts a full-night single-channel EEG signal as input and produces segmentation masks for arousal and sleep stage labels. FullSleepNet comprises four modules: a convolutional module to extract local features, a recurrent module to capture long-range dependencies, an attention mechanism to focus on relevant parts of the input, and a segmentation module to output final predictions. Main results. By unifying the two interrelated tasks as segmentation problems and employing a multi-task learning approach, FullSleepNet achieves state-of-the-art performance for arousal detection with an area under the precision-recall curve of 0.70 on Sleep Heart Health Study and Multi-Ethnic Study of Atherosclerosis datasets. For sleep stage classification, FullSleepNet obtains comparable performance on both datasets, achieving an accuracy of 0.88 and an F1-score of 0.80 on the former and an accuracy of 0.83 and an F1-score of 0.76 on the latter. Significance. Our results demonstrate that FullSleepNet offers improved practicality, efficiency, and accuracy for the detection of arousal and classification of sleep stages using raw EEG signals as input.
Segmantasyon yapmadan patolojik kalp sesi kayıtlarının tespiti için bir örüntü sınıflandırma algoritması
(DÜMF Mühendislik Dergisi, 2019) Yıldız, Abdulnasır; Zan, Hasan
Bu çalışmada, altı adet veri bankasından alınan kalp sesi kayıtlarına segmentasyon uygulamadan k-En Küçük Komşuluk (kNN), Destek Vektör Makinesi (DVM) ve sınıflandırıcı metotları topluluğu kullanarak sınıflandırmaya yarayan bir algoritma geliştirilmesi amaçlanmıştır. Altı aşamadan oluşan algoritmanın ilk aşaması olan Önişlem aşamasında sinyaller sivri uçlarından arındırılmış ve ardından normalize edilmiştir. İkinci aşama olan Özellik çıkarma–1 aşamasında, sinyalin çeşitli zaman ve frekans özellikleri çıkarılarak üçüncü aşamadaki veri bankası sınıflandırıcının eğitilmesinde ve test edilmesinde kullanılmıştır. Üçüncü aşama olan veri bankası sınıflandırması aşamasında, her veri bankası oluşturulurken kullanılan cihazların, kayıt yeri ve ortamının farklılığının negatif etkilerini azaltmak ve her veri bankası için farklı özellikler ile sınıflandırıcılar kullanmak amacıyla kalp sesi kayıtları veri bankalarına göre sınıflandırılmıştır. Dördüncü aşama olan Özellik Çıkarma-2 aşamasında veri bankalarına göre sınıflandırılan sinyallerin yine çeşitli zaman ve zaman-frekans özellikleri çıkarılmıştır. Beşinci aşamada her veri bankası için 3 farklı sınıflandırıcı (kNN, DVM ve sınıflandırıcı topluluğu) kullanılarak kayıtlar sınıflandırılmıştır. Algoritmanın son aşaması olan Oylama aşamasında, nihai sınıflandırma başarımını arttırmak amacıyla her kayıt için 3 farklı sınıflandırıcının çıkışları belli kurallara göre oylanarak kaydın sınıfı (patolojik veya normal) belirlenmiştir. Beşli çapraz doğrulama kullanılarak eğitilen ve test (tanı testi) edilen algoritmanın performansı ölçülürken doğruluk, duyarlılık, özgüllük, pozitif ve negatif yorum gücü ile ROC grafiğinin altında kalan alan gibi parametreler kullanılmıştır. En iyi performans sonuçları doğruluk: %94.28, duyarlılık: %87.97, özgüllük: %87.97, pozitif yorum gücü: %84.78, negatif yorum gücü: %96.86 ve ROC eğrisi altında kalan alan: 0.919 şeklinde elde edildi. Elde edilen bu değerler daha önceki çalışmalar ile kıyaslandığında algoritmanın oldukça başarılı olduğu ve kalbin patolojik durumuna ilişkin uzman hekime ön tanı imkânı sunabileceği söylenebilir.
Sleep arousal detection using one dimensional local binary pattern-based convolutional neural network
(IEEE, 2021) Zan, Hasan; Yıldız, Abdulnasır; Zan, Hasan
Sleep arousal is defined as a shift from deep sleep to light sleep or complete awakening. Arousals cause sleep deprivation by fragmenting sleep, and ultimately, many health problems. Arousals can be induced by well-studied apneas and hypopneas or other sleep orders such as hypoventilation, bruxism, respiratory effort-related arousals. Thus, detection of less-studied non-apnea/hypopnea arousals is important for diagnosis and treatment of sleep disorders. Traditionally, polysomnography (PSG) test that is recording and inspecting overnight physiological signals is used for sleep studies. In this work, a novel method based on one dimensional local binary pattern (1D-LBP) and convolutional neural network (CNN) for automatic arousal detection from polysomnography recordings is proposed. 25 recordings from PhysioNet Challenge 2018 PSG dataset are used for experiments. Each signal in PSG recordings is transformed to a new signal using 1D-LBP, and then segmented using 10-s-long sliding window. The segments are fed to a CNN model formed by stacking 25 layers for classification of non-apnea/hypopnea arousal regions from non-arousal regions. Area under precision-recall curve (AUPRC) and area under receiver operating characteristic curve (AUROC) metrics are used for performance measurement. Experimental results reflect that the proposed method shows a great promise and obtains an AUPRC of 0.934 and an AUROC of 0.866.