Automatic seizure detection using three-dimensional CNN based on multi-channel EEG

Background

Automated seizure detection from clinical EEG data can reduce the diagnosis time and facilitate targeting treatment for epileptic patients. However, current detection approaches mainly rely on limited features manually designed by domain experts, which are inflexible for the detection of a variety of patterns in a large amount of patients’ EEG data. Moreover, conventional machine learning algorithms for seizure detection cannot accommodate multi-channel Electroencephalogram (EEG) data effectively, which contains both temporal and spatial information. Recently, deep learning technology has been widely applied to perform image processing tasks, which could learns useful features from data and process multi-channel data automatically. To provide an effective system for automatic seizure detection, we proposed a new three-dimensional (3D) convolutional neural network (CNN) structure, whose inputs are multi-channel EEG signals.

Methods

EEG data of 13 patients were collected from one center hospital, which has already been inspected by experts. To represent EEG data in CNN, firstly time series of each channel of EEG data was converted into the two-dimensional image. Then all channel images were combined into 3D images according to the mutual correlation intensity between different electrodes. Finally, a CNN was constructed using 3D kernels to predict different stages of EEG data, including inter-ictal, pre-ictal, and ictal stages. The system performance was evaluated and compared with the traditional feature-based classifier and the two-dimensional (2D) deep learning method.

Results

It demonstrated that multi-channel EEG data could provide more information for increasing the specificity and sensitivity in cpmparison result between the single and multi-channel. And the 3D CNN based on multi-channel outperformed the 2D CNN and traditional signal processing methods with an accuracy of more than 90%, an sensitivity of 88.90% and an specificity of 93.78%.

Conclusions

This is the first effort to apply 3D CNN in detecting seizures from EEG. It provides a new way of learning patterns simultaneously from multi-channel EEG signals, and demonstrates that deep neural networks in combination with 3D kernels can establish an effective system for seizure detection.

     

Hidden Markov model based epileptic seizure detection using tunable Q wavelet transform

Epilepsy is one of the most prevalent neurological disorders affecting 70 million people worldwide. The present work is focused on designing an efficient algorithm for automatic seizure detection by using electroencephalogram (EEG) as a noninvasive procedure to record neuronal activities in the brain. EEG signals' underlying dynamics are extracted to differentiate healthy and seizure EEG signals. Shannon entropy, collision entropy, transfer entropy, conditional probability, and Hjorth parameter features are extracted from subbands of tunable Q wavelet transform. Efficient decomposition level for different feature vector is selected using the Kruskal-Wallis test to achieve good classification. Different features are combined using the discriminant correlation analysis fusion technique to form a single fused feature vector. The accuracy of the proposed approach is higher for Q=2 and J=10. Transfer entropy is observed to be significant for different class combinations. Proposed approach achieved 100% accuracy in classifying healthy-seizure EEG signal using simple and robust features and hidden Markov model with less computation time. The proposed approach efficiency is evaluated in classifying seizure and non-seizure surface EEG signals. The system has achieved 96.87% accuracy in classifying surface seizure and nonseizure EEG segments using efficient features extracted from different J level.     

Application of adaptive neuro-fuzzy inference system for epileptic seizure detection using wavelet feature extraction

Intelligent computing tools such as artificial neural network (ANN) and fuzzy logic approaches are demonstrated to be competent when applied individually to a variety of problems. Recently, there has been a growing interest in combining both these approaches, and as a result, neuro-fuzzy computing techniques have been evolved. In this study, a new approach based on an adaptive neuro-fuzzy inference system (ANFIS) was presented for epileptic seizure detection. The proposed ANFIS model combined the neural network adaptive capabilities and the fuzzy logic qualitative approach. Decision making was performed in two stages: feature extraction using the wavelet transform (WT) and the ANFIS trained with the backpropagation gradient descent method in combination with the least squares method. Some conclusions concerning the impacts of features on the detection of epileptic seizures were obtained through analysis of the ANFIS. The results are highly promising, and a comparative analysis suggests that the proposed modeling approach outperforms ANN model in terms of training performances and classification accuracies. The results confirmed that the proposed ANFIS model has some potential in epileptic seizure detection. The ANFIS model achieved accuracy rates which were higher than that of the stand-alone neural network model.     

Automatic seizure detection with different time delays using SDFT and time-domain feature extraction

Automatic seizure detection is important for fast detection of the seizure because the way that the expert denotes and searches for seizure in the long signal takes time. The most common way to detect seizures automatically is to use an electroencephalogram(EEG). Many studies have used feature extraction that needs time for calculation. In this study, sliding discrete Fourier transform(SDFT) was applied for conversion to a frequency domain without using a window, which was compared with using wi...     

EEG reactivity in high and low symptomatic schizophrenics,using source modelling in the frequency domain

Summary A dipole localization method in the frequency domain was used (FFT Dipole Approximation) to assess spatial differences in the spectral EEG reactivity (orienting response) between high and low symptomatic schizophrenics. Frequency bands of interest were determined empirically by comparing the two dichotomized patient groups with two matched control groups. Evidence for a correlation between EEG reactivity and severity of schizophrenic symptomatology was found, especially in the higher beta frequency range (16 – 25.5 Hz). Opposite effects were found in the two beta ranges of 20.5–22.5 Hz and 23.0–25.5 Hz, supporting the hypothesis that different EEG frequency bands have specific functional significances and that these bands are not necessarily those that are conventionally selected.We thank W. Manske and E. Tremel for their collaboration in conducting the recordings and initial steps of data treatment, and Dr. R. D. Pascual-Marqui for his contribution to the statistical analysis of the data.     

Quantification of MRS data in the frequency domain using a wavelet filter, an approximated Voigt lineshape model and prior knowledge

Quantification of MRS spectra is a challenging problem when a large baseline is present along with a low signal to noise ratio. This work investigates a robust fitting technique that yields accurate peak areas under these conditions. Using simulated long echo time (1)H MRS spectra with low signal to noise ratio and a large baseline component, both the accuracy and reliability of the fit in the frequency domain were greatly improved by reducing the number of fitted parameters and making full use of all the known information concerning the Voigt lineshape. Using an appropriate first order approximation to a popular approximation of the Voigt lineshape, a significant improvement in the estimate of the area of a known spectral peak was obtained with a corresponding reduction in the residual. Furthermore, this improved parameter choice resulted in a large reduction in the number of iterations of the least-squares fitting routine. On the other hand, making use of the known centre frequency differences of the component resonances gave negligible improvement. A wavelet filter was used to remove the baseline component. In addition to performing a Monte Carlo study, these fitting techniques were also applied to a set of 10 spectra acquired from healthy human volunteers. Again, the same reduced parameter model gave the lowest value for chi(2) in each case.     

Automated human mind reading using EEG signals for seizure detection

Abstract

Epilepsy is one of the most occurring neurological disease globally emerged back in 4000 BC. It is affecting around 50 million people of all ages these days. The trait of this disease is recurrent seizures. In the past few decades, the treatments available for seizure control have improved a lot with the advancements in the field of medical science and technology. Electroencephalogram (EEG) is a widely used technique for monitoring the brain activity and widely popular for seizure region detection. It is performed before surgery and also to predict seizure at the time operation which is useful in neuro stimulation device. But in most of cases visual examination is done by neurologist in order to detect and classify patterns of the disease but this requires a lot of pre-domain knowledge and experience. This all in turns put a pressure on neurosurgeons and leads to time wastage and also reduce their accuracy and efficiency. There is a need of some automated systems in arena of information technology like use of neural networks in deep learning which can assist neurologists. In the present paper, a model is proposed to give an accuracy of 98.33% which can be used for development of automated systems. The developed system will significantly help neurologists in their performance.     

Differential operator in seizure detection

Differential operators can detect significant changes in signals. This has been utilized to enhance the contrast of the seizure signatures in depth EEG or ECoG. We have actually taken normalized exponential of absolute value of single or double derivative of epileptic ECoG. This in short we call differential filtering. Windowed variance operation has been performed to automatically detect seizure onset on differentially filtered signal. A novel method for determining the duration of seizure has also been proposed. Since all operations take only linear time, the whole method is extremely fast. Seven empirical parameters have been introduced whose patient specific thresholding brings down the rate of false detection to a bare minimum. Results of implementation of the methods on the ECoG data of four epileptic patients have been reported with an ROC curve analysis. High value of the area under the ROC curve indicates excellent detection performance.     

Epileptic seizure detection using EEG signals and extreme gradient boosting

The problem of automated seizure detection is treated using clinical electroencephalograms (EEG) and machine learning algorithms on the Temple University Hospital EEG Seizure Corpus (TUSZ). Performances on this complex data set are still not encountering expectations. The purpose of this work is to determine to what extent the use of larger amount of data can help to improve the performances. Two methods are explored: a standard partitioning on a recent and larger version of the TUSZ, and a leave-one-out approach used to increase the amount of data for the training set. XGBoost, a fast implementation of the gradient boosting classifier, is the ideal algorithm for these tasks. The performances obtained are in the range of what is reported until now in the literature with deep learning models. We give interpretation to our results by identifying the most relevant features and analyzing performances by seizure types. We show that generalized seizures tend to be far better predicted than focal ones. We also notice that some EEG channels and features are more important than others to distinguish seizure from background.     

EEG epileptic seizure detection and classification based on dual-tree complex wavelet transform and machine learning algorithms

The visual analysis of common neurological disorders such as epileptic seizures in electroencephalography (EEG) is an oversensitive operation and prone to errors, which has motivated the researchers to develop effective automated seizure detection methods. This paper proposes a robust automatic seizure detection method that can establish a veritable diagnosis of these diseases. The proposed method consists of three steps: (i) remove artifact from EEG data using Savitzky-Golay filter and multi-scale principal component analysis (MSPCA), (ii) extract features from EEG signals using signal decomposition representations based on empirical mode decomposition (EMD), discrete wavelet transform (DWT), and dual-tree complex wavelet transform (DTCWT) allowing to overcome the non-linearity and non-stationary of EEG signals, and (iii) allocate the feature vector to the relevant class (i.e., seizure class "ictal" or free seizure class "interictal") using machine learning techniques such as support vector machine (SVM), k-nearest neighbor (k-NN), and linear discriminant analysis (LDA). The experimental results were based on two EEG datasets generated from the CHB-MIT database with and without overlapping process. The results obtained have shown the effectiveness of the proposed method that allows achieving a higher classification accuracy rate up to 100% and also outperforms similar state-of-the-art methods.     

基于高频颅内脑电的致痫区定位和网络分析

目的通过癫痫发作期颅内高频脑电(60~90 Hz)的分析,实现一种基于非参数统计模型的致痫区定位方法,并研究致痫区内外的脑电信号相干性。方法首先构造高频脑电归一化指标——高频能量指数,再用该指标量化各导联的致痫性,并定位致痫区,最后分析致痫区内外电极的幅度相干和相位相干性。结果分析了8例有良好术后效果的癫痫患者共1252个电极触点的颅内脑电数据,以临床癫痫灶定位结果作为参考,高频能量指数平均敏感性29.57%、特异性96.67%。8位患者的致痫电极之间的平均相干性显著高于非致痫电极之间的相干性。结论构造了颅内脑电高频能量指标,并用非参数统计模型定位致痫区,特异性高于已有研究,能够帮助医生在癫痫手术规划中寻找关键相关致痫电极。     

The detection of impedance cardiogram characteristic points using wavelet transform

A novel impedance cardiograph event detection method using wavelet transform is proposed. When compared to the C and E points in the pressure-volume loop, the wavelet method performs significantly better than the traditional method (P < 0.05) in the B and X points detection even after the addition of 20% artificial noise into the test signal. Nevertheless, the SVs estimated by ICG are poorly correlated with values measured by the conductance catheter.     

Automatic gene annotation using GO terms from cellular component domain

Background

The Gene Ontology (GO) is a resource that supplies information about gene product function using ontologies to represent biological knowledge. These ontologies cover three domains: Cellular Component (CC), Molecular Function (MF), and Biological Process (BP). GO annotation is a process which assigns gene functional information using GO terms to relevant genes in the literature. It is a common task among the Model Organism Database (MOD) groups. Manual GO annotation relies on human curators assigning gene functional information using GO terms by reading the biomedical literature. This process is very time-consuming and labor-intensive. As a result, many MODs can afford to curate only a fraction of relevant articles.

Methods

GO terms from the CC domain can be essentially divided into two sub-hierarchies: subcellular location terms, and protein complex terms. We cast the task of gene annotation using GO terms from the CC domain as relation extraction between gene and other entities: (1) extract cases where a protein is found to be in a subcellular location, and (2) extract cases where a protein is a subunit of a protein complex. For each relation extraction task, we use an approach based on triggers and syntactic dependencies to extract the desired relations among entities.

Results

We tested our approach on the BC4GO test set, a publicly available corpus for GO annotation. Our approach obtains a F1-score of 71%, a precision of 91% and a recall of 58% for predicting GO terms from CC Domain for given genes.

Conclusions

We have described a novel approach of treating gene annotation with GO terms from CC domain as two relation extraction subtasks. Evaluation results show that our approach achieves a F1-score of 71% for predicting GO terms for given genes. Thereby our approach can be used to accelerate the process of GO annotation for the bio-annotators.

     

Effect of filtering with time domain features for the detection of epileptic seizure from EEG signals

Pattern recognition plays an important role in the detection of epileptic seizure from electroencephalogram (EEG) signals. In this pattern recognition study, the effect of filtering with the time domain (TD) features in the detection of epileptic signal has been studied using naive Bayes (NB) and supports vector machines (SVM). It is the first time the authors attempted to use TD features such as waveform length (WL), number of zero-crossings (ZC) and number of slope sign changes (SSC) derived from the filtered and unfiltered EEG data, and performance of these features is studied along with mean absolute value (MAV) which has been already attempted by the researchers. The other TD features which are attempted by the researchers such as standard deviation (SD) and average power (AVP) along with MAV are studied. A comparison is made in effect of filtering and without filtering for the University of Bonn database using NB and SVM for the TD features attempted first time along with MAV. The effect of individual and combined TD features is studied and the highest classification accuracy obtained in using direct TD features would be 99.87%, whereas it is 100% with filtered EEG data. The raw EEG data can be segmented and filtered using the fourth-order Butterworth band-pass filter.     

ECG analysis using wavelet transform: application to myocardial ischemia detection

In this paper, we propose a method for the detection of myocardial ischemic events from electrocardiogram (ECG) signal using the wavelet transform technique. The wavelet transform is obtained using the quadratic spline wavelet. Then, based on the wavelet transform values, the characteristic points of the ECG signal are found out. These characteristic points are used to identify any ischemic episodes present in the ECG signal. This technique can be extended for other types of cardiac abnormality detections, which induce changes in the ECG.     

Wavelength optimization for rapid chromophore mapping using spatial frequency domain imaging

Spatial frequency-domain imaging (SFDI) utilizes multiple-frequency structured illumination and model-based computation to generate two-dimensional maps of tissue absorption and scattering properties. SFDI absorption data are measured at multiple wavelengths and used to fit for the tissue concentration of intrinsic chromophores in each pixel. This is done with a priori knowledge of the basis spectra of common tissue chromophores, such as oxyhemoglobin (ctO(2)Hb), deoxyhemoglobin (ctHHb), water (ctH(2)O), and bulk lipid. The quality of in vivo SFDI fits for the hemoglobin parameters ctO(2)Hb and ctHHb is dependent on wavelength selection, fitting parameters, and acquisition rate. The latter is critical because SFDI acquisition time is up to six times longer than planar two-wavelength multispectral imaging due to projection of multiple-frequency spatial patterns. Thus, motion artifact during in vivo measurements compromises the quality of the reconstruction. Optimal wavelength selection is examined through matrix decomposition of basis spectra, simulation of data, and dynamic in vivo measurements of a human forearm during cuff occlusion. Fitting parameters that minimize cross-talk from additional tissue chromophores, such as water and lipid, are determined. On the basis of this work, a wavelength pair of 670 nm∕850 nm is determined to be the optimal two-wavelength combination for in vivo hemodynamic tissue measurements provided that assumptions for water and lipid fractions are made in the fitting process. In our SFDI case study, wavelength optimization reduces acquisition time over 30-fold to 1.5s compared to 50s for a full 34-wavelength acquisition. The wavelength optimization enables dynamic imaging of arterial occlusions with improved spatial resolution due to reduction of motion artifacts.     

Simulation of ultrasound two-dimensional array transducers using a frequency domain model

Ultrasound imaging with two-dimensional (2D) arrays has garnered broad interest from scanner manufacturers and researchers for real time three-dimensional (3D) applications. Previously the authors described a frequency domain B-mode imaging model applicable for linear and phased array transducers. In this paper, the authors extend this model to incorporate 2D array transducers. Further approximations can be made based on the fact that the dimensions of the 2D array element are small. The model is compared with the widely used ultrasound simulation program FIELD II, which utilizes an approximate form of the time domain impulse response function. In a typical application, errors in simulated RF waveforms are less than 4% regardless of the steering angle for distances greater than 2 cm, yet computation times are on the order of 1/35 of those incurred using FIELD II. The 2D model takes into account the effects of frequency-dependent attenuation, backscattering, and dispersion. Modern beam-forming techniques such as apodization, dynamic aperture, dynamic receive focusing, and 3D beam steering can also be simulated.     

Hair detection and lesion segmentation in dermoscopic images using domain knowledge

Automated segmentation and dermoscopic hair detection are one of the significant challenges in computer-aided diagnosis (CAD) of melanocytic lesions. Additionally, due to the presence of artifacts and variation in skin texture and smooth lesion boundaries, the accuracy of such methods gets hampered. The objective of this research is to develop an automated hair detection and lesion segmentation algorithm using lesion-specific properties to improve the accuracy. The aforementioned objective is achieved in two ways. Firstly, a novel hair detection algorithm is designed by considering the properties of dermoscopic hair. Second, a novel chroma-based geometric deformable model is used to effectively differentiate the lesion from the surrounding skin. The speed function incorporates the chrominance properties of the lesion to stop evolution at the lesion boundary. Automatic initialization of the initial contour and chrominance-based speed function aids in providing robust and flexible segmentation. The proposed approach is tested on 200 images from PH2 and 900 images from ISBI 2016 datasets. Average accuracy, sensitivity, specificity, and overlap scores of 93.4, 87.6, 95.3, and 11.52% respectively are obtained for the PH2 dataset. Similarly, the proposed method resulted in average accuracy, sensitivity, specificity, and overlap scores of 94.6, 82.4, 97.2, and 7.20% respectively for the ISBI 2016 dataset. Statistical and quantitative analyses prove the reliability of the algorithm for incorporation in CAD systems.

Open image in new window

Graphical Abstract Overview of proposed system

     

Brain mapping and detection of functional patterns in fMRI using wavelet transform; application in detection of dyslexia

Background

Functional Magnetic Resonance Imaging (fMRI) has been proven to be useful for studying brain functions. However, due to the existence of noise and distortion, mapping between the fMRI signal and the actual neural activity is difficult. Because of the difficulty, differential pattern analysis of fMRI brain images for healthy and diseased cases is regarded as an important research topic. From fMRI scans, increased blood ows can be identified as activated brain regions. Also, based on the multi-sliced images of the volume data, fMRI provides the functional information for detecting and analyzing different parts of the brain.

Methods

In this paper, the capability of a hierarchical method that performed an optimization algorithm based on modified maximum model (MCM) in our previous study is evaluated. The optimization algorithm is designed by adopting modified maximum correlation model (MCM) to detect active regions that contain significant responses. Specifically, in the study, the optimization algorithm is examined based on two groups of datasets, dyslexia and healthy subjects to verify the ability of the algorithm that enhances the quality of signal activities in the interested regions of the brain. After verifying the algorithm, discrete wavelet transform (DWT) is applied to identify the difference between healthy and dyslexia subjects.

Results

We successfully showed that our optimization algorithm improves the fMRI signal activity for both healthy and dyslexia subjects. In addition, we found that DWT based features can identify the difference between healthy and dyslexia subjects.

Conclusion

The results of this study provide insights of associations of functional abnormalities in dyslexic subjects that may be helpful for neurobiological identification from healthy subject.