Search

Empirical Mode Decomposition (EMD) provides an adaptive signal processing tool, and its multivariate extension is useful to model multichannel signals. Recently, EMD and multivariate EMD have succe...

Dynamic mode decomposition (DMD) is a new matrix decomposition method proposed as an iterative solution to problems in fluid flow analysis. Recently, DMD algorithm has successfully been applied to the analysis of non-stationary signals such as neural recordings. In this study, we propose single-channel, and multi-channel EEG based DMD approaches for the analysis of epileptic EEG signals. We investigate the possibility of utilizing the “DMD Spectrum” for the classification of pre-seizure and seizure EEG segments. We introduce higher-order DMD spectral moments and DMD sub-band powers, and extract them as features for the classification of epileptic EEG signals. Experiments are conducted on multi-channel EEG signals collected from 16 epilepsy patients. Single-channel, and multi-channel EEG based DMD approaches have been tested on epileptic EEG data recorded from only right, only left, and both brain hemisphere channels. Performance of various classifiers using the proposed DMD-Spectral based features are compared with that of traditional spectral features. Experimental results reveal that the higher order DMD spectral moments and DMD sub-band power features introduced in this study, outperform the analogous spectral features calculated from traditional power spectrum.

In this study, a liking estimation system based on electroencephalogram (EEG) signals is developed for neuromarketing applications. The determination of the degree of appreciation of a product by consumers has become an important research topic using machine learning methods. Biological data is recorded while viewing product pictures or videos, then processed by signal processing methods. In this study, 32 channel EEG signals are recorded from subjects who watched two different car advertisement videos and the liking status is determined. After watching the advertisement videos, the participants were asked to vote for the rating of the different images (front view, dashboard, side view, rear view, taillight, logo and grille) of the products. The signals corresponding to these different video regions from the EEG recordings were segmented and analyzed by the Empirical Mode Decomposition (EMD) and Ensemble Empirical Mode Decomposition (EEMD). The statistical features were extracted from Intrinsic Mode Functions (IMF) and the liking status classifications were performed. The classification performance of EMD- and EEMD-based methods are 93.4% and 97.8% respectively on Brand1, and 93.5% and 97.4% respectively on Brand2. In addition, the classification accuracy on both brands combined are 85.1% and 85.7% respectively. The promising results obtained using Support Vector Machines (SVM) show that the proposed EEG-based method may be used in neuromarketing studies.

Background Epilepsy is one of the most common neurological disorders associated with disruption of brain activity. In the classification and detection of epileptic seizures, electroencephalography (EEG) measurements, which record the electrical activities of the brain, are frequently used. Empirical mode decomposition (EMD) and its derivative, ensemble EMD (EEMD) are recently developed methods used to decompose non-stationary and nonlinear signals such as EEG into a finite number of oscillations called intrinsic mode functions (IMFs). Our main objective in this study is to present a hybrid IMF selection method combining four different approaches (energy, correlation, power spectral distance, and statistical significance measures), and investigate the effect of selected IMFs extracted by EMD and EEMD on the classification. We have applied the proposed IMF selection approach on the classification of EEG signals recorded from epilepsy patients who are under treatment at our collaborator hospital. Multichannel EEG signals collected from epilepsy patients are decomposed into IMFs, and then IMF selection was performed. Finally, time- and spectral-domain, and nonlinear features are extracted and feature sets are created for the classification. Results The maximum classification accuracies obtained using various combinations of IMFs were 94.56%, 95.63%, 96.8%, and 96.25% for SVM, KNN, naive Bayes, and logistic regression classifiers, respectively, by using EMD analysis; whereas, the EEMD approach has provided maximum classification accuracies of 96.06%, 97%, 97%, and 96.25% for SVM, KNN, naive Bayes, and logistic regression, respectively. Classification performance with the same features obtained using direct EEG signals instead of the decomposed IMFs was worse than the aforementioned 2 approaches for every combination. Conclusion Simulation results demonstrate that the proposed IMF selection approach affects the classification results. Also, EEMD provides a robust method for feature extraction from EEG signals in order to classify pre-seizure and seizure segments.

Epilepsy is a persistent and recurring neurological condition in a community of brain neurons that results from sudden and abnormal electrical discharges. This paper introduces a new form of assessment and interpretation of the changes in electroencephalography (EEG) recordings from different brain regions in epilepsy disorders based on graph analysis and statistical rescale range analysis. In this study, two different states of epilepsy EEG data (preictal and ictal phases), obtained from 17 subjects (18 channels each), were analyzed by a new method called state transfer network (STN). The analysis performed by STN yields a network metric called motifs, which are averaged over all channels and subjects in terms of their persistence level in the network. The results showed an increase of overall motif persistence during the ictal over the preictal phase, reflecting the synchronization increase during the seizure phase (ictal). An evaluation of intermotif cross-correlation indicated a definite manifestation of such synchronization. Moreover, these findings are compared with several other well-known methods such as synchronization likelihood (SL), visibility graph similarity (VGS), and global field synchronization (GFS). It is hinted that the STN method is in good agreement with approaches in the literature and more efficient. The most significant contribution of this research is introducing a novel nonlinear analysis technique of generalized synchronization. The STN method can be used for classifying epileptic seizures based on the synchronization changes between multichannel data.

This research presents a new method for detecting obsessive–compulsive disorder (OCD) based on time–frequency analysis of multi-channel electroencephalogram (EEG) signals using the multi-variate synchrosqueezing transform (MSST). With the evolution of multi-channel sensor implementations, the employment of multi-channel techniques for the extraction of features arising from multi-channel dependency and mono-channel characteristics has become common. MSST has recently been proposed as a method for modeling the combined oscillatory mechanisms of multi-channel signals. It makes use of the concepts of instantaneous frequency (IF) and bandwidth. Electrophysiological data, like other nonstationary signals, necessitates both joint time–frequency analysis and independent time and frequency domain studies. The usefulness and effectiveness of a multi-variate, wavelet-based synchrosqueezing algorithm paired with a band extraction method are tested using electroencephalography data obtained from OCD patients and control groups in this research. The proposed methodology yields substantial results when analyzing differences between patient and control groups.

This paper presents a novel method for emotion recognition based on time-frequency analysis using multivariate synchrosqueezing transform (MSST) of multichannel electroencephalography (EEG) signals. With the advancements of the multichannel sensor applications, the need for multivariate algorithms has become obvious for extracting features that stem from multichannel dependency in addition to mono-channel features. In order to model the joint oscillatory structure of these multichannel signals, MSST has recently been proposed. It uses the concepts of joint instantaneous frequency and bandwidth. Electrophysiological data processing mostly requires joint time-frequency analysis in addition to both time and frequency analysis separately. The short-time Fourier transform (STFT) and wavelet transform (WT) are the main approaches utilized in time-frequency analysis. In this paper, the feasibility and performance of multivariate wavelet-based synchrosqueezing algorithm was demonstrated on EEG signals obtained from publically available DEAP database by comparing with its univariate version. Eight emotional states were considered by combining arousal-valence and dominance dimensions. Using linear support vector machines (SVM) as a classifier, MSST and its univariate version resulted in the highest prediction accuracy rates of ˜93% among all emotional states.

Epilepsy is one of the most common brain disorders worldwide. The most frequently used clinical tool to detect epileptic events and monitor epilepsy patients is the EEG recordings. There have been proposed many computer-aided diagnosis systems using EEG signals for the detection and prediction of seizures. In this study, a novel method based on Fourier-based Synchrosqueezing Transform (SST), which is a high-resolution time-frequency (TF) representation, and Convolutional Neural Network (CNN) is proposed to detect and predict seizure segments. SST is based on the reassignment of signal components in the TF plane which provides highly localized TF energy distributions. Epileptic seizures cause sudden energy discharges which are well represented in the TF plane by using the SST method. The proposed SST-based CNN method is evaluated using the IKCU dataset we collected, and the publicly available CHB-MIT dataset. Experimental results demonstrate that the proposed approach yields high average segment-based seizure detection precision and accuracy rates for both datasets (IKCU: 98.99% PRE and 99.06% ACC; CHB-MIT: 99.81% PRE and 99.63% ACC). Additionally, SST-based CNN approach provides significantly higher segment-based seizure prediction performance with 98.54% PRE and 97.92% ACC than similar approaches presented in the literature using the CHB-MIT dataset.

Background: Electrocardiogram (ECG) is a method of recording the electrical activity of the heart and it provides a diagnostic means for heart-related diseases. Arrhythmia is any irregularity of the heartbeat that causes an abnormality in the heart rhythm. Early detection of arrhythmia has great importance to prevent many diseases. Manual analysis of ECG recordings is not practical for quickly identifying arrhythmias that may cause sudden deaths. Hence, many studies have been presented to develop computer-aided-diagnosis (CAD) systems to automatically identify arrhythmias.Methods: This paper proposes a novel deep learning approach to identify arrhythmias in ECG signals. The proposed approach identifies arrhythmia classes using Convolutional Neural Network (CNN) trained by two-dimensional (2D) ECG beat images. Firstly, ECG signals, which consist of 5 different arrhythmias, are segmented into heartbeats which are transformed into 2D grayscale images. Afterward, the images are used as input for training a new CNN architecture to classify heartbeats. Results: The experimental results show that the classification performance of the proposed approach reaches an overall accuracy of 99.7%, sensitivity of 99.7%, and specificity of 99.22% in the classification of five different ECG arrhythmias. Further, the proposed CNN architecture is compared to other popular CNN architectures such as LeNet and ResNet-50 to evaluate the performance of the study.Conclusions: Test results demonstrate that the deep network trained by ECG images provides outstanding classification performance of arrhythmic ECG signals and outperforms similar network architectures. Moreover, the proposed method has lower computational costs compared to existing methods and is more suitable for mobile device-based diagnosis systems as it does not involve any complex preprocessing process. Hence, the proposed approach provides a simple and robust automatic cardiac arrhythmia detection scheme for the classification of ECG arrhythmias.

Electroencephalography (EEG) signals are frequently used for the detection of epileptic seizures. In this chapter, advanced signal analysis methods such as Empirical Mode Decomposition (EMD), Ensembe (EMD), Dynamic mode decomposition (DMD), and Synchrosqueezing Transform (SST) are utilized to classify epileptic EEG signals. EMD and its derivative, EEMD are recently developed methods used to decompose nonstationary and nonlinear signals such as EEG into a finite number of oscillations called intrinsic mode functions (IMFs). In this study multichannel EEG signals collected from epilepsy patients are decomposed into IMFs, and then essential IMFs are selected. Finally, time- and spectral-domain, and nonlinear features are extracted from selected IMFs and classified. DMD is a new matrix decomposition method proposed as an iterative solution to problems in fluid flow analysis. We present single-channel, and multi-channel EEG based DMD approaches for the analysis of epileptic EEG signals. As a third method, we use the SST representations of seizure and pre-seizure EEG data. Various features are calculated and classified by Support Vector Machine (SVM), k-Nearest Neighbor (kNN), Naive Bayes (NB), Logistic Regression (LR), Boosted Trees (BT), and Subspace kNN (S-kNN) to detect pre-seizure and seizure signals. Simulation results demonstrate that the proposed approaches achieve outstanding validation accuracy rates.

Epilepsy is a neurological disease that is very common worldwide. In the literature, patient’s electroencephalography (EEG) signals are frequently used for an epilepsy diagnosis. However, the success of epileptic examination procedures from quantitative EEG signals is limited. In this paper, a high-resolution time-frequency (TF) representation called Synchrosqueezed Transform (SST) is used to classify epileptic EEG signals. The SST matrices of seizure and pre-seizure EEG data of 16 epilepsy patients are calculated. Two approaches based on machine learning and deep learning are proposed to classify pre-seizure and seizure signals. In the machine learning-based approach, the various features like higher-order joint moments are calculated and these features are classified by Support Vector Machine (SVM), k-Nearest Neighbor (kNN) and Naive Bayes (NB) classifiers. In the deep learning-based approach, the SST matrix was recorded as an image and a Convolutional Neural Network (CNN)-based architecture was used to classify these images. Simulation results demonstrate that both approaches achieved promising validation accuracy rates. While the maximum (90.2%) validation accuracy is achieved for the machine learning-based approach, (90.3%) validation accuracy is achieved for the deep learning-based approach.

Epilepsy is a neurological disorder that affects many people all around the world, and its early detection is a topic of research widely studied in signal processing community. In this paper, a new technique that was introduced to solve problems of fluid dynamics called Dynamic Mode Decomposition (DMD), is used to classify seizure and non-seizure epileptic EEG signals. The DMD decomposes a given signal into the intrinsic oscillations called modes which are used to define a DMD spectrum. In the proposed approach, the DMD spectrum is obtained by applying either multi-channel or single-channel based DMD technique. Then, subband and total power features extracted from the DMD spectrum and various classifiers are utilized to classify seizure and non-seizure epileptic EEG segments. Outstanding classification results are achieved by both the single-channel based (96.7%), and the multi-channel based (96%) DMD approaches.

Epilepsy is a type of neurological disorder that causes abnormal brain activities and creates epileptic seizures. Traditionally epileptic seizure prediction is realized with a visual examination of Electroencephalogram (EEG) signals. But this technique needs a long time EEG monitoring. So, the automatic epileptic seizures prediction schemes become a requirement at this point. This study proposes a method to classify epileptic seizures and normal EEG data by utilizing the Intrinsic Time-scale Decomposition (ITD)-based features. The dataset has been supplied from the database of the Epileptology Department of Bonn University. It contains 5 data groups A, B, C, D, E. The study aims to classify healthy and epileptic data, so data of groups A and E are used to perform evaluations of proposed methods. The EEG data are decomposed into Proper Rotation Components (PRCs) by ITD. The feature extraction methods are applied to the first five PRCs of each EEG data from healthy and epileptic individuals. These features are classified using K-Nearest Neighbors (KNN), Linear Discriminant Analysis (LDA), Naive Bayes, Support Vector Machine (SVM) and Logistic Regression classifiers. The results demonstrated that the epileptic data is differentiated from normal data by applying the nonlinear ITD with outstanding classification performance.

The Electromyography (EMG) signal is a nonstationary bio-signal based on the measurement of the electrical activity of the muscles. EMG based recognition systems play an important role in many fields such as diagnosis of neuromuscular diseases, human-computer interactions, console games, sign language detection, virtual reality applications, and amputee device controls. In this study, a novel approach based on deep learning has been proposed to improve the accuracy rate in the prediction of hand movements. Firstly, 4-channel surface EMG (sEMG) signals have been measured while simulating 7 different hand gestures (Extension, Flexion, Open Hand, Punch, Radial Deviation, Rest, and Ulnar Deviation) from 30 participants. The obtained sEMG signals have been segmented into sections where each movement was found. Then, spectrogram images of the segmented sEMG signals have been created by means of ShortTime Fourier Transform (STFT). The created colored spectrogram images have trained with 50-layer Convolutional Neural Network (CNN) based on Residual Networks (ResNet) architecture. Owing to the proposed method, test accuracy of 99.59% and F1 Score of 99.57% have achieved for 7 different hand gesture classifications.

In this study, it was aimed to recognize the emotional state from facial images using the deep learning method. In the study, which was approved by the ethics committee, a custom data set was created using videos taken from 20 male and 20 female participants while simulating 7 different facial expressions (happy, sad, surprised, angry, disgusted, scared, and neutral). Firstly, obtained videos were divided into image frames, and then face images were segmented using the Haar library from image frames. The size of the custom data set obtained after the image preprocessing is more than 25 thousand images. The proposed convolutional neural network (CNN) architecture which is mimics of LeNet architecture has been trained with this custom dataset. According to the proposed CNN architecture experiment results, the training loss was found as 0.0115, the training accuracy was found as 99.62%, the validation loss was 0.0109, and the validation accuracy was 99.71%.

In this study, it is determined individual-based features which are used to estimate emotional negative valence and compared the features effectiveness with different classifiers. Ten movie clips are shown to subjects as an emotional stimuli and EEG recording is recorded synchronously. Emotional valence value is scored in [–7 7] Likert scale by the subjects immediately after video ended. According to lowest and highest valence values, two classes are generated. The data is processed on an individual basis and personal spatial filters is obtained by Independent Component Analysis. After calculating the spectrogram of the spatial filtered data, features are extracted by subtracting amplitudes of 3Hz averaged frequency bands. The result of feature selection, it is observed that features from beta and gamma bands are much more effective. The success rate of the selected features was tested with five classifiers by cross validation, and high performance was obtained from multilayer perceptron classifiers and the instance- based k-nearest neighborhood algorithm (IBk-NN). The average accuracies of IBk-NN and multilayer classifier are achieved 86% ±8 and 83% ±9, respectively.

The heart is the most critical organ for the sustainability of life. Arrhythmia is any irregularity of heart rate that causes an abnormality in your heart rhythm. Clinical analysis of Electrocardiogram (ECG) signals is not enough to quickly identify abnormalities in the heart rhythm. This paper proposes a deep learning method for the accurate detection of abnormal and normal heartbeats based on 2-D Convolutional Neural Network (CNN) architecture. Two channels of ECG signals were obtained from the MIT-BIH arrhythmia dataset. Each ECG signal is segmented into heartbeats, and each heartbeat is transformed into a 2-D grayscale heartbeat image as an input for CNN structure. Due to the success of image recognition, CNN architecture is utilized for binary classification of the 2-D image matrix. In this study, the effect of different CNN architectures is compared based on the classification rate. The accuracies of training and test data are found as 100.00% and 99.10%, respectively for the best CNN model. Experimental results demonstrate that CNN with ECG image representation yields the highest success rate for the binary classification of ECG beats compared to the traditional machine learning methods, and one-dimensional deep learning classifiers.

Computer systems working with artificial intelligence can recognize movements and gestures to be used for many purposes. In order to perform recognition, the electrical activity of the muscles can be utilized which is represented by electromyography (EMG) and EMG is not a stationary biological signal. EMG based movement recognition systems have an important place in distinct areas like in human-computer interactions, virtual reality, prosthesis, and hand exoskeletons. In this study, a new approach based on deep learning (DL) and Empirical Mode Decomposition (EMD) is proposed to improve the accuracy rate for recognition of hand movements in its application areas. Firstly, 4-channel surface EMG (sEMG) signals were measured while simulating 7 different hand gestures, which are extension, flexion, ulnar deviation, radial deviation, punch, open hand, and rest, from 30 subjects. After that, noiseless signals were procured utilizing filters as a result of preprocessing. Then, pre-processed signals were subjected to segmentation. Thereafter, the EMD process was applied to each segmented signal and Intrinsic Mode Functions (IMFs) were obtained. The IMFs time-series which are some kind of screen images of the first 3 IMFs have been recorded. For classification, IMFs images have given as inputs and have trained to the 101layer Convolution Neural Network (CNN) based on Residual Networks (ResNet) architecture, which is a DL model.

Time-frequency representation (TFR) provides a good analysis for periodic signals; however, they are insufficient for nonstationary signals. The synchrosqueezing transform (SST) provides a strong analysis of nonstationary signals. The signal has different synchrosqueezing transformations that are implemented using different TFR. This paper provides a review of the different SST methods implemented using different TFR available in the literature, a comparison of these, and their use with different techniques in biomedical signal processing applications. Adding different techniques to the applied SST method affects the signal processing and classification ability of the selected SST method.

In this study, an estimation system based on electroencephalogram (EEG) signals has been developed for use in neuromarketing applications. Determination of the degree of consumer liking a product by processing the biological data (EEG, facial expressions, eye tracking, Galvanic skin response, etc.) recorded while viewing the product images or videos has become an important research topic. In this study, 32-channel EEG signals were recorded from subjects while they watch two different car advertisement videos, and the liking status was determined. After watching the car commercial videos, the subjects were asked to vote on the rating of different images (front view, front console, side view, rear view, rear light, logo and front grill) of the cars. The signals corresponding to these different video regions from the EEG recordings were segmented and analyzed by the Empirical Mode Decomposition (EMD) method. Several statistical features were extracted from the resulting Intrinsic Mode Functions and the liking status classification was performed. Classification results obtained with Support Vector Machines (SVM) classifiers indicate that the proposed EEG-based liking detection method may be used in neuromarketing studies.

In the literature, several signal processing techniques have been used to diagnose epilepsy which is a nervous system disease. However most of these techniques fail to analyse EEG signals which are dynamic and non-linear. In this study, an approach which utilizes a data-driven technique called Dynamic Mode Decomposition (DMD) that was originally developed to be used in fluid mechanics was proposed. Features that were belonged to EEG signals were calculated using DMD method and with the help of different classifiers, classification of the preseizure and seizure EEG signals was performed. Obtained results showed that the proposed method presented an alternative to approaches that are based on Empirical Mode Decomposition and its derivatives.

Arrhythmia is any irregularity of heart rate that cause an abnormality in your heart rhythm. Manual analysis of Electrocardiogram (ECG) signal is not enough for quickly identify abnormalities in the heart rhythm. This paper proposes a deep learning approach for detection of five different arrhythmia types based on 2D convolutional neural networks (CNN) architecture. ECG signals were obtained from MIT-BIH arrhythmia database. For CNN architecture, each ECG signal was segmented into heartbeats, then each heartbeat was transformed into 2D grayscale heartbeat image. 2D CNN model was used due to success of image recognition. The proposed model result demonstrate that CNN and ECG image formation give highest result when classified different types of ECG arrhythmic signals.

Emotion recognition is an effective analysis method used to increase the interaction between human-machine interface. EEG based emotion recognition studies based on brain signals are preferred in order to provide healthy results of emotion analysis experiments. In this study, emotion recognition analysis was performed in accordance with dimensional emotion modelling. Data cleaning was performed by applying the necessary filters on the recorded data. The feature vector was then created and the success rate was determined using support vector machines and classification methods such as K-nearest neighbour.

Obsessive–compulsive disorder (OCD) is one of the neuropsychiatric disorders qualified by intrusive and iterative annoying thoughts and mental attitudes that are activated by these thoughts. In recent studies, advanced signal processing techniques have been favored to diagnose OCD. This research suggests four different measurements; intrinsic phase-locked value, intrinsic coherence, intrinsic synchronization likelihood, and intrinsic visibility graph similarity that quantifies the synchronization level and complexity in electroencephalography (EEG) signals. This intrinsic synchronization is achieved by utilizing Multivariate Empirical Mode Decomposition (MEMD), a data-driven method that resolves nonlinear and nonstationary data into their intrinsic mode functions. Our intrinsic technique in this study demonstrates that MEMD-based synchronization analysis gives us much more detailed knowledge rather than utilizing the synchronization method alone. Furthermore, the nonlinear synchronization method presents more consistent results considering OCD heterogeneity. Statistical evaluation using sample [Formula: see text]-test and [Formula: see text]-test has shown the significance of such new methodology.

Background: Electrocardiogram (ECG) is a method of recording the electrical activity of the heart and provides a diagnostic mean for heart-related diseases. An arrhythmia is any irregularity of heartbeat that causes an abnormality in one’s heart rhythm. Early detection of arrhythmia has great importance to prevent many diseases. Manual analysis of ECG signal is not sufficient for quickly identifying arrhythmias that may cause sudden deaths. Hence, many studies have been presented to developed computer-aided diagnosis (CAD) systems to automatically identify arrhythmias.Methods: This paper proposes a novel deep learning approach to identify arrhythmias in ECG signals. The signals are obtained from MIT-BIH arrhythmia database and are categorized according to five arrhythmia types. The proposed approach identifies arrhythmia classes by using Convolutional Neural Network (CNN) architecture trained by two-dimensional (2D) ECG beat images. CNN architecture is selected due to high image recognition performance. ECG signals are segmented into heartbeats, then each heartbeat is transformed into a 2D grayscale image. The heartbeat images are used as input for the CNN. Results: The proposed CNN model is compared to other common CNN architectures such as LeNet and ResNet-50 to evaluate the performance of our study. Overall, the proposed study achieved 99.7% test accuracy in the classification of five different ECG arrhythmias.Conclusions: Testing results demonstrate that CNN trained by ECG image representations provide outstanding classification performance of arrhythmic ECG signals and outperforms similar network architectures. Hence, the proposed approach provides a robust method for the classification of ECG arrhythmias.

The emotional state of people plays a key role in physiological and behavioral human interaction. Emotional state analysis entails many fields such as neuroscience, cognitive sciences, and biomedical engineering because the parameters of interest contain the complex neuronal activities of the brain. Electroencephalogram (EEG) signals are processed to communicate brain signals with external systems and make predictions over emotional states. This paper proposes a novel method for emotion recognition based on deep convolutional neural networks (CNNs) that are used to classify Valence, Arousal, Dominance, and Liking emotional states. Hence, a novel approach is proposed for emotion recognition with time series of multi-channel EEG signals from a Database for Emotion Analysis and Using Physiological Signals (DEAP). We propose a new approach to emotional state estimation utilizing CNN-based classification of multi-spectral topology images obtained from EEG signals. In contrast to most of the EEG-based approaches that eliminate spatial information of EEG signals, converting EEG signals into a sequence of multi-spectral topology images, temporal, spectral, and spatial information of EEG signals are preserved. The deep recurrent convolutional network is trained to learn important representations from a sequence of three-channel topographical images. We have achieved test accuracy of 90.62% for negative and positive Valence, 86.13% for high and low Arousal, 88.48% for high and low Dominance, and finally 86.23% for like–unlike. The evaluations of this method on emotion recognition problem revealed significant improvements in the classification accuracy when compared with other studies using deep neural networks (DNNs) and one-dimensional CNNs.

This paper investigates the feasibility of using time–frequency (TF) representation of EEG signals for emotional state recognition. A recent and advanced TF analyzing method, multivariate synchrosqueezing transform (MSST) is adopted as a feature extraction method due to multi-channel signal processing and compact component localization capabilities. First, the 32 participants' EEG recordings from DEAP emotional EEG database are analyzed using MSST to reveal oscillations. Second, independent component analysis (ICA), and feature selection are applied to reduce the high dimensional 2D TF distribution without losing distinctive component information in the 2D image. Thus, only one method for feature extraction using MSST is performed to analyze time, and frequency-domain properties of the EEG signals instead of using some signal analyzing combinations (e.g., power spectral density, energy in bands, Hjorth parameters, statistical values, and time differences etc.). As well, the TF-domain reduction performance of ICA is compared to non-negative matrix factorization (NMF) to discuss the accuracy levels of high/low arousal, and high/low valence emotional state recognition. The proposed MSST-ICA feature extraction approach yields up to correct rates of 82.11%, and 82.03% for arousal, and valence state recognition using artificial neural network. The performances of the MSST and ICA are compared with Wigner-Ville distribution (WVD) and NMF to investigate the effects of TF distributions as feature set with reduction techniques on emotion recognition.

DOI: 10.26650/electrica.2018.00998 Emotion state detection or emotion recognition cuts across different disciplines because of the many parameters that embrace the brain's complex neural structure, signal processing methods, and pattern recognition algorithms. Currently, in addition to classical time-frequency methods, emotional state data have been processed via data-driven methods such as empirical mode decomposition (EMD). Despite its various benefits, EMD has several drawbacks: it is intended for univariate data; it is prone to mode mixing; and the number of local extrema must be enough before the EMD process can begin. To overcome these problems, this study employs a multivariate EMD and its noise-assisted version in the emotional state classification of electroencephalogram signals.

This paper explores the data-driven properties of the empirical mode decomposition (EMD) for detection of epileptic seizures. A new method in frequency domain is presented to analyze intrinsic mode functions (IMFs) decomposed by EMD. They are used to determine whether the electroencephalogram (EEG) recordings contain seizure or not. Energy levels of the IMFs are extracted as threshold level to detect the changes caused by seizure activity. A scalar value energy resulting from the energy levels is individually used as an indicator of the epileptic EEG without the requirements of multidimensional feature vector and complex machine learning algorithms. The proposed methods are tested on different EEG recordings to evaluate the effectiveness of the proposed method and yield accuracy rate up to 97.89%.

Most real-life signals exhibit non-stationary characteristics. Processing of such signals separately in the time-domain or in the frequency-domain does not provide sufficient information as their spectral properties change over time. Traditional methods such as the Fourier transform (FT) perform a transformation from time-domain to frequency-domain allowing a suitable spectral analysis but looses the spatial/temporal information of the signal components. Hence, it is not easy to observe a direct relationship between the time and frequency characteristics of the signal. This makes it difficult to extract useful information by using only time- or frequency-domain analysis for further processing purposes. To overcome this problem, joint time–frequency (TF) methods are developed and applied to the analysis and representation of non-stationary signals. In addition to revealing a time-dependent energy distribution information, TF methods have successfully been utilized in the estimation of some parameters related to the analyzed signals. In this paper, we briefly summarize the existing methods and present several state-of-the-art applications of TF methods in the classification of biomedical signals. We also point out some future perspectives for the processing of non-stationary signals in the joint TF domain.

Emotions play a significant role in daily life by encouraging the individual in the survival, decision making, guessing, and communication processes. Through emotions can be explained with the activation of anatomical structures in certain regions of brain with nervous system the emotions can be understood by electroencephalogram (EEG) signals. In order to recognize emotions, the signal processing techniques were applied to recorded signals using 32-channels EEG device from the subjects during listening audios. The Self-Assessment Manikin (SAM) form was filled by 23 subjects to evaluate their feelings based on three emotion states and recorded their answers by designed Graphical User Interface (GUI) monitored in front of the subjects. In signal processing stage, the EEG signals were segmented into segmented files by cutting stimulus intervals from recorded signal and decomposed to Intrinsic Mode Functions (IMFs) by Empirical Mode Decomposition (EMD) method. Then, most meaningful IMFs has been selected by analyzing Power Spectral Density (PSD) to extract statistical and entropy-based features and then, classification algorithm has been applied to obtain feature vector to categorize states of emotion consisting of valence, arousal, dominance dimensions. It is aimed to find most useful selected IMFs, most active channels related to emotion, best suitable features for each dimension of emotion. Finally, the percentage of performance accuracy has been calculated and the best accuracy of 81.74% is found in channels ranged in frontal lobe (1–12) for valence state, 72.15% in channel TP7 for arousal state, 74.57% in channels ranged in frontal lobe (1–12) for dominance state by combining different features.

Empirical mode decomposition (EMD) method is a powerful tool in the analysis of of nonlinear and nonstationary signals. It decomposes signals into a number of amplitude and frequency modulated signals namely intrinsic mode functions (IMFs). However, some of these IMFs represents the original signal better while some of them are useless. IMF selection methods are suggested to determine the IMFs which represents the original signal better than other IMFs. In this study, we analyzed the effect of IMF selection methods in classification performance. We compared power based, correlation based and power spectral density based IMF selection methods in the classification of the electroencephalography (EEG) signals, which are collected from subjects with multiple sclerosis. The EEG signals are classified as the patients are being cognitively impaired or intact. k-nearest neighbors, multilayer perceptron neural networks and random forest classifiers are used in classification. The results show that, effect of IMF selection methods on accuracy is changeable in regard to classifier preference.

Counting of microbial colonies is crucial due to the applications of medical microbiology to search and detect the causes of diseases. While different tasks performed, the counting process of bacteria colonies is provided either by the searcher manually or by a common software, nowadays. The manual counting of bacteria colonies is tiresome, eye-straining, and time-consuming for the searcher where common softwares require high troublesome with having high error rates. The aim of this study is detecting and counting bacteria colonies without having these limitations in today's non-practical applications. Therefore, an image-processing based bacteria colony counter designed in MATLAB. In the medical plasma laboratory of the Izmir Katip Celebi University three different types of hospital-acquired infection cause bacterias, which are Escherichia coli, Pseudomonas aeruginosa, and Enterococcus faecalis, cultured and examined properly, then, using the Circular Hough Transform (CHT) in MATLAB the detection and counting of bacteria colonies provided. To be able to obtain more practical usage, a Graphical User Interface (GUI) designed.

© 2019 IEEE.Arrhythmia is irregular changes of normal heart rhythm and effective manual identifying of them require a lot of time and depends on experience of clinicians. This paper proposes deep learning-based novel 2-D convolutional neural network (CNN) approach for accurate classification of five different arrhythmia types. The performance of the proposed architecture is tested on Electrocardiogram (ECG) signals that are taken from MIT-BIH arrhythmia benchmark database. ECG signals was segmented into heartbeats and each of the heartbeats was converted into 2-D grayscale images as an input data for CNN structure. The accuracy of the proposed architecture was found as 97.42% on the training results revealed that the proposed 2-D CNN architecture with transformed 2-D ECG images can achieve highest accuracy without any preprocessing and feature extraction and feature selection stages for ECG signals.

Emotion detection is very crucial role on diagnosis of brain disorders and psychological disorders. Electroencephalogram (EEG) is useful tool that obtain complex physiological brain signals from human. In this paper, we proposed a novel approach for emotional state estimation using convolutional neural network (CNN) based deep learning technique from EEG signals. Firstly, we convert 32 lead EEG signals to 2D EEG images with Azimuthal Equidistant Projection (AEP) technique. Then, 2D images that represented measurements of EEG signals sent to CNN based deep neural network for classification. In this study, we have achieved accuracy of 95.96% two classes as negative and positive valence, 96.09% two classes as high and low arousal and 95.90% two classes as high and low arousal dominance.

In this study, it is aimed to select the Intrinsic mode functions that can best distinguish pre-seizure and seizure segments of epileptic EEG signals by using the Intrinsic mode functions (IMF) obtained by Ensemble Empirical Mode Decomposition (EEMD) method. In our study, a hybrid method was proposed based on various IMF selection methods, and the first 3 IMFs were found to have the highest priority. In order to determine the contribution of IMF selection to the classification accuracy, various spectral features were calculated and the classification was performed by using Support Vector Machines, Naive Bayes, K-Nearest Neighbor, and Linear Discriminant Analysis methods. Upon checking the classification results obtained using the first 3 IMFs, it is observed that the classification accuracy is higher with the features obtained using first MF which was found to have the highest priority at the IMF selection process.

We propose a technique for training convolutional neural networks (CNNs) in which the convolutional layers are trained using a gradient descent based method and the classification layer is trained using a second order method called approximate sparse multinomial logistic regression (ASMLR) which also provides a spatial smoothing procedure that increases the classification accuracy for hyperspectral images. ASMLR performs well on hyperspectral images, and CNNs are known to give good results in many applications such as image classification and object recognition. Thus, the proposed technique allows us to improve the performance of CNNs by training the whole network with an end-to-end framework. This approach takes advantage of convolutional layers for spectral feature extraction, and of the softmax classification layer for feature selection with sparsity constraints, and an intrinsic learning rate adjustment mechanism. In classification, we also use a spatial smoothing method. The proposed method was evaluated on two hyperspectral images for spectral-spatial land cover classification, and the results have shown that it outperforms the CNN and the ASMLR classifiers when they are used separately.

Three main requirements of a successful application of deep learning are the network architecture, a large enough training dataset, and a good optimization algorithm. In this paper we mainly focus on the optimization part. We propose a training algorithm for convolutional neural networks which makes use of both first and second order derivatives for training different layers. We utilize an approximate second order algorithm for the classification layer while we train the rest of the network with the conventional approach which is backpropagation with first order derivatives. We show that this approach helps us achieve a higher classification accuracy with a much smaller number of training iterations compared to training the whole network with gradient descent based algorithms. Moreover, although second order optimization is generally costlier, we show that the proposed approach is trained faster not only in terms of the number of iterations but also training duration. We also present the integration of CNNs with a probabilistic spatial model and apply this to the land cover classification problem in hyperspectral images. The results show that the algorithm allows us to achieve superior results with a simple network even with limited training data compared to existing approaches.

Epilepsy is a neurological disease that is very common worldwide. Patient’s electroencephalography (EEG) signals are frequently used for the detection of epileptic seizure segments. In this paper, a high-resolution time-frequency (TF) representation called Synchrosqueezing Transform (SST) is used to detect epileptic seizures. Two different EEG data sets, the IKCU data set we collected, and the publicly available CHB-MIT data set are analyzed to test the performance of the proposed model in seizure detection. The SST representations of seizure and nonseizure (pre-seizure or inter-seizure) EEG segments of epilepsy patients are calculated. Various features like higher-order joint TF (HOJ-TF) moments and gray-level co-occurrence matrix (GLCM)-based features are calculated using the SST representation. By using single and ensemble machine learning methods such as k-Nearest Neighbor (kNN), Logistic Regression (LR), Naive Bayes (NB), Support Vector Machine (SVM), Boosted Trees (BT), and Subspace kNN (S-kNN), EEG features are classified. The proposed SST-based approach achieved 95.1% ACC, 96.87% PRE, 95.54% REC values for the IKCU data set, and 95.13% ACC, 93.37% PRE, 90.30% REC values for the CHB-MIT data set in seizure detection. Results show that the proposed SST-based method utilizing novel TF features outperforms the short-time Fourier transform (STFT)-based approach, providing over 95% accuracy for most cases, and compares well with the existing methods.

© 2018 IEEE.One of the main causes of sudden deaths is heart disease. Early detection and treatment of cardiac arrhythmias prevent the problem from reaching sudden deaths. The purpose of this study is to develop an arrhythmia detection algorithm based on Empirical Mode Decomposition (EMD). This algorithm consists of four steps: Preprocessing, Empirical Mode Decomposition, feature extraction and classification. Six arrhythmia types were used for differentiate normal and arrhythmic signals obtained from the MIT-BIH Arrhythmia database. These are normal (N), left bundle branch block (LBBB), right bundle branch block (RBBB), premature ventricular contraction (PVC), paced beat and atrial premature beats (APB). Three different classifiers were used to classify ECG signals. The method achieves better result with accuracy of 87% using linear discriminant analysis (LDA) classifier for detection of normal and arrhythmic signals.

Epilepsy is a neurological disorder of different types characterized by recurrent of seizures which affects people of all ages. This paper presents visibility graph similarity as a nonlinear model to analyze the epilepsy EEG data from different brain region of healthy and patient subjects with epilepsy seizures. All EEG segments are mapped into a corresponding graph to obtain the corresponding degree of sequence for each segment, and then the difference between these degrees is constructed as a similarity between two segments. The results showed that seizure activity presented strongest nonlinear dynamic response in the form of similarity level decreasing from healthy subjects to patients. Results of other sets were found to be in agreement with our results.

This study investigates improved properties of empirical mode decomposition (EMD) for emotion recognition by using electroencephalogram (EEG) signals. The emotion recognition from EEG signals is a difficult study by the reason of nonstationary behavior of the signals. These signals are affected from complicated neural activity of brain. To analyze EEG signals, advanced signal processing techniques are required. In our study, data are collected from one channeled BIOPAC lab system. EEG signals were obtained from visual evoked potentials of 13 female and 13 male volunteers for 12 pleasant and 12 unpleasant pictures. To analyze nonlinear and nonstationary characteristics of EEG signals, an EMD-based method is proposed for emotion recognition. Various time and frequency domain techniques such as power spectral density (PSD), and higher order statistics (HOS) are used to analyze the IMFs extracted by EMD. Support vector machine (SVM), Linear discriminant analysis (LDA), and Naive Bayes classifiers are utilized for the classification of features extracted from the IMFs, and their performances are compared.

In recent years, utilizing Hilbert-based time- frequency methods in emotional state sensing research attracted attention in the brain computer interfaces. Primarily, Hilbert Transform-based empirical mode decomposition (EMD) was found to be suitable for emotional state modeling studies. In more recent studies, models of emotional state recognition were proposed in which the classification was implemented by using the features obtained after applying the time, frequency, and time-frequency domain methods to intrinsic mode functions achieved by operating EMD. In this study, an analysis of emotional state recognition is proposed by using the features of the synchrosqueezing coefficients obtained in the classification process after applying the Synchrosqueezing Transform to intrinsic mode functions achieved by using Multivariate EMD. As a result, EEG data available in the DEAP database were categorized as low and high for valence, activation, and dominance dimensions, and 4 different classifiers were utilized in the classification process. The most satisfying ratios of valence, activation and dominance were attained 76%, 68%, and 68% respectively.

The detection of epileptic seizure has a primary role in patient diagnosis with epilepsy. Epilepsy causes sudden and uncontrolled electrical discharges in brain cells. Recordings of the abnormal brain activities are time consuming and outcomes are very subjective, so automated detection systems are highly recommended. In this study, it is aimed to classify EEG signals for the detection of epileptic seizures using intrinsic mode functions (IMF) and feature extraction based on Empirical Mode Decomposition (EMD). These records have been acquired from the database of the Epileptology Department of Bonn University and consisting of 5 marker groups A, B, C, D, E in this study. These records taken from healthy individuals and patients are decomposed into IMFs by EMD method. Feature vectors have been extracted based on Tsallis Entropy, Renyi Entropy, Relative Entropy and Coherence methods. These features are then classified by K-Nearest Neighbors Classification (KNN), Linear Discriminant Analysis (LDA) and Naive Bayes Classification (NBC). Significant differences were determined between healthy and patient EEG data at the end of the study.

EEG is one of the most used devices in brain examinations, provides important information about many neurological diseases. The power of EEG signal's phase synchronization is a parameter that has been commonly used in brain investigations. The Interchannels Phase Clustering (ICPC) method that we propose in our study, is an improved version of the Intertrial Phase Clustering (ICPC) method. Global and regional phase synchronization studies can be done with ICPC method. The method is applied to the signals produced by an EEG simulator and the results are computed with high accuracy. This method can be applied to neuropsychiatric diseases and provide important information on functional connectivity and cognitive functions.

Emotion detection by utilizing signal processing methods is a challenging area. An open issue in emotional modeling is to obtain an optimum feature set to use for the classification process. This study proposes an approach for emotional state classification by the investigation of EEG signals via multivariate synchrosqueezing transform (MSST). MSST is a post-processing technique to compose a localized time-frequency representation yielding multivariate syncyrosqueezing coefficients. After obtaining these coefficients from EEG signals for 18 subjects from DEAP dataset, coefficients and self-assessment-mannequins (SAM) labels of those subjects are used for emotional state classification by using support vector machines (SVM) nearest neighbor, decision tree, and ensemble methods. The accuracy rate is 70.6% for high valence high arousal (HVHA), 75.4% for low valence high arousal (LVHA), 77.8% for high valence low arousal (HVLA), and 77.2% for low valence low arousal (LVLA) cases using SVM.

Attention deficit hyperactivity disorder (ADHD) is a psychiatric condition that affects millions of children and many times last into adulthood. There is no single test that can show whether a person has ADHD. The symptoms vary from person to person. Therefore, it is hard to diagnose ADHD contrary to many physical illnesses. Our aim is to create methods to minimize human effort and increase accuracy of diagnosis of ADHD. We collected structural Magnetic Resonance Images (MRI) from 26 subjects: 11 controls and 15 children diagnosed with ADHD. The data was provided from NPIstanbul NeuroPsychiatric Hospital. We developed automatic, effective, rapid, and accurate framework for diagnosing ADHD. The models were built on the k-nearest neighbors algorithm (KNN) and naive Bayes using Matlab machine learning toolbox. Shape and texture feature extraction technique was used. Area, Perimeter, Eccentricity, EquivDiameter, Major Axis Length, Minor Axis Length, Orientation are characteristics used for shape feature extraction technique. Textural features of a magnetic resonance imaging were represented with first (mean, variance, skewness, kurtosis) and second order statistical (contrast, correlation, homogeneity, energy) based feature extraction techniques. Gray and white regions were extracted using k-means algorithms. Local features were extracted from these regions by shape and texture methods. Global features were extracted with second order statistics which is called gray level co-occurrence method. The most important attribute was determined by using principal component analysis. The experiments were conducted on a full training dataset including 26 instance and 5 fold cross validation was adopted for randomly sampling training and test sets. ADHD is successfully classified with 100 % accuracy by using the proposed method. The outcome of our study will reduce the number medical errors by informing physicians in their efforts of diagnosing ADHD.

Epilepsy is a chronic neurological disorder affects people of all ages; this paper presents cross-correlation and state transfer network method to analyze synchronization between EEG data-channels from subjects with epilepsy seizures. The datasets are in two phases, preictal and ictal phase. All EEG segments are mapped into a corresponding state network graph to obtain the corresponding motifs and then the cross-correlation is applied to exhibit the synchronization changing during epilepsy seizures. The results showed that ictal phase presented high synchronization between channels, where low synchronization level is observed within preictal phase.

Electroencephalogram (EEG) signals are widely used to investigate the electrical activities in the brain an to get information for these activities. In this work, samples are selected from 3 type (Negative, Neutral, Positive) emotional pictures from ‘Nencki Affective Picture System’ and shown to subjects. EEG signals are collected according to 10–20 international electrode system standards. Collected signals are reduced to 3 type channels (Frontal, Parieto-Occipital, Central). Power Spectral Density (PSD) of the corresponding EEG signals are investigated and separated into EEG frequency bands (Delta, Theta, Alpha, Beta). Welch Spectral Estimation Method is applied into the acquired data and their PSDs are obtained after taking the expectation of their frequencies. Using PSDs, total power is calculated for each frequency band and channel. Then ANOVA analysis is performed and statistical differences are investigated. Our analyses show that statistically significant differences exist between each frequency band and 3 type of channels and among reduced channels themselves.

In this study a probabilistic method for contextual classification of hyperspectral images is proposed with the purpose of land cover classification. The proposed method consists of a multinomial logistic regression model, and multi-nomial autologistic regression model for spatial smoothing. The parameters in this model are approximately calculated by a lower bound method. Simulation results show that the proposed contextual hyperspectral image classification method yields high classification accuracies.