Adaptive spatio-temporal filtering of multichannel surface EMG signals

A motor unit (MU) is defined as an anterior horn cell, its axon, and the muscle fibres innervated by the motor neuron. A surface electromyogram (EMG) is a superposition of many different MU action potentials (MUAPs) generated by active MUs. The objectives of this study were to introduce a new adaptive spatio-temporal filter, here called maximum kurtosis filter (MKF), and to compare it with existing filters, on its performance to detect a single MUAP train from multichannel surface EMG signals. The MKF adaptively chooses the filter coefficients by maximising the kurtosis of the output. The proposed method was compared with five commonly used spatial filters, the weighted low-pass differential filter (WLPD) and the marginal distribution of a continuous wavelet transform. The performance was evaluated using simulated EMG signals. In addition, results from a multichannel surface EMG measurement fro from a subject who had been previously exposed to radiation due to cancer were used to demonstrate an application of the method. With five time lags of the MKF, the sensitivity was 98.7% and the highest sensitivity of the traditional filters was 86.8%, which was obtained with the WLPD. The positive predictivities of these filters were 87.4 and 80.4%, respectively. Results from simulations showed that the proposed spatio-temporal filtration technique significantly improved performance as compared with existing filters, and the sensitivity and the positive predictivity increased with an increase in number of time lags in the filter.     

MUAP extraction and classification based on wavelet transform and ICA for EMG decomposition

We have developed an effective technique for extracting and classifying motor unit action potentials (MUAPs) for electromyography (EMG) signal decomposition. This technique is based on single-channel and short periodȁ9s real recordings from normal subjects and artificially generated recordings. This EMG signal decomposition technique has several distinctive characteristics compared with the former decomposition methods: (1) it bandpass filters the EMG signal through wavelet filter and utilizes threshold estimation calculated in wavelet transform for noise reduction in EMG signals to detect MUAPs before amplitude single threshold filtering; (2) it removes the power interference component from EMG recordings by combining independent component analysis (ICA) and wavelet filtering method together; (3) the similarity measure for MUAP clustering is based on the variance of the error normalized with the sum of RMS values for segments; (4) it finally uses ICA method to subtract all accurately classified MUAP spikes from original EMG signals. The technique of our EMG signal decomposition is fast and robust, which has been evaluated through synthetic EMG signals and real EMG signals.     

Filtering of surface EMG using ensemble empirical mode decomposition

Surface electromyogram (EMG) is often corrupted by three types of noises, i.e. power line interference (PLI), white Gaussian noise (WGN), and baseline wandering (BW). A novel framework based primarily on empirical mode decomposition (EMD) was developed to reduce all the three noise contaminations from surface EMG. In addition to regular EMD, the ensemble EMD (EEMD) was also examined for surface EMG denoising. The advantages of the EMD based methods were demonstrated by comparing them with the traditional digital filters, using signals derived from our routine electrode array surface EMG recordings. The experimental results demonstrated that the EMD based methods achieved better performance than the conventional digital filters, especially when the signal to noise ratio of the processed signal was low. Among all the examined methods, the EEMD based approach achieved the best surface EMG denoising performance.     

Simulation Studies of Latency Measures of Components of the Event-Related Brain Potential

We compared the accuracy of P300 latency estimates obtained with different procedures under several simulated signal and noise conditions. Both preparatory and signal detection techniques were used. Preparatory techniques included frequency filters and spatial filters (single electrode selection and Vector filter). Signal detection techniques included peak-picking, cross-correlation, and Woody filter. Accuracy in the latency estimation increased exponentially as a function of the signal-to-noise ratio. Both Woody filter and cross-correlation provided better estimates than peak-picking, although this advantage was reduced by frequency filtering. For all signal detection techniques, Vector filter provided better estimates than single electrode selection. Large component overlap impaired the accuracy of the estimates obtained with both single electrode selection and Vector filter, but with Vector filter impairment occurred only when the overlapping component had a scalp distribution that was similar to the scalp distribution of the signal component. The effects of varying noise characteristics, P300 duration and latency, and the parameters of Vector filter were also investigated.     

Effect of spatial filtering on crosstalk reduction in surface EMG recordings

Increasing the selectivity of the detection system in surface electromyography (EMG) is beneficial in the collection of information of a specific portion of the investigated muscle and to reduce the contribution of undesired components, such as non-propagating components (due to generation or end-of-fibre effects) or crosstalk from nearby muscles. A comparison of the ability of different spatial filters to reduce the amount of crosstalk in surface EMG measurements was conducted in this paper using simulated signals. It focused on the influence of different properties of the muscle anatomy (changing subcutaneous layer thickness, skin conductivity, fibre length) and detection system (single, double and normal double differential, with two inter-electrode distances - IED) on the amount of crosstalk present in the measurements. A cylindrical multilayer (skin, subcutaneous tissue, muscle, bone) analytical model was used to simulate single fibre action potentials (SFAPs). Fibres were grouped together in motor units (MUs) and motor unit action potentials (MUAPs) were obtained by adding the SFAPs of the corresponding fibres. Interference surface EMG signals were obtained, modelling the recruitment of MUs and rate coding. The average rectified value (ARV) and mean frequency (MNF) content of the EMG signals were studied and used as a basis for determining the selectivity of each spatial filter. From these results it was found that the selectivity of each spatial filter varies depending on the transversal location of the measurement electrodes and on the anatomy. An increase in skin conductivity favourably affects the selectivity of normal double differential filters as does an increase in subcutaneous layer thickness. An increase in IED decreases the selectivity of all the analysed filters.     

Real-time ocular artifact suppression using recurrent neural network for electro-encephalogram based brain-computer interface

The paper presents an adaptive noise canceller (ANC) filter using an artificial neural network for real-time removal of electro-oculogram (EOG) interference from electro-encephalogram (EEG) signals. Conventional ANC filters are based on linear models of interference. Such linear models provide poorer prediction for biomedical signals. In this work, a recurrent neural network was employed for modelling the interference signals. The eye movement and eye blink artifacts were recorded by the placing of an electrode on the forehead above the left eye and an electrode on the left temple. The reference signal was then generated by the data collected from the forehead electrode being added to data recorded from the temple electrode. The reference signal was also contaminated by the EEG. To reduce the EEG interference, the reference signal was first low-pass filtered by a moving averaged filter and then applied to the ANC. Matlab Simulink was used for real-time data acquisition, filtering and ocular artifact suppression. Simulation results show the validity and effectiveness of the technique with different signal-to-noise ratios (SNRs) of the primary signal. On average, a significant improvement in SNR up to 27 dB was achieved with the recurrent neural network. The results from real data demonstrate that the proposed scheme removes ocular artifacts from contaminated EEG signals and is suitable for real-time and short-time EEG recordings.     

Single motor unit analysis from spatially filtered surface electromyogram signals. Part I: Spatial selectivity

The aim of the study was to compare experimentally, on the basis of single motor unit (MU) activities, the selectivity of different spatial filters commonly used to detect surface electromyogram (EMG) signals. Surface EMG signals were recorded from the biceps brachii and the upper trapezius muscle of five subjects using a two-dimensional (2D) electrode array consisting of 16 pin electrodes. The subjects performed isometric contractions at different elbow angles and shoulder abduction and flexion. The same monopolar surface EMG signals were filtered using longitudinal single and double differential, transverse single and double differential and normal double differential filters. From the single MU action potentials, extracted by automatic EMG decomposition, indexes of transverse (perpendicular with respect to the fibre direction) and longitudinal (along the fibre direction) selectivity were computed. The number of detected MUs was 48 for the upper trapezius, with the arms held in the sagittal plane, and 52 when the arms were held in the frontal plane; 85 MUs were identified from the biceps brachii contractions. The results showed that transverse selectivity was significantly higher for the 2D and transverse onedimensional (1D) filters with respect to the 1D longitudinal filters, whereas longitudinal selectivity was higher (i.e. MU action potentials were shorter) for the 2D filter and the longitudinal double differential filter. In particular, the relative attenuation of potential amplitude moving 5 mm from the source was, on average (for the two muscles), 16.5% for the least selective filter in the transverse direction (longitudinal single differential) and 35.7% for the most selective one in the same direction (transverse double differential). The MU action potential duration was, on average, 13.8 ms for the most selective filter in the longitudinal direction (longitudinal double differential) and 18.7 ms for the least selective one (transverse double differential). The normal double differential filter resulted in spatial selectivity indexes that were not statistically different in the two directions from those of the best filters in each direction.     

Use of finite wordlength FIR digital filter structures with improved magnitude and phase characteristics for reduction of muscle noise in EEG signals

One of the main disturbances in EEG signals is EMG artefacts generated by muscle movements. In the paper, the use of a linear phase FIR digital low-pass filter with finite wordlength precision coefficients is proposed, designed using the compensation procedure, to minimise EMG artefacts in contaminated EEG signals. To make the filtering more effective, different structures are used, i.e. cascading, twicing and sharpening (apart from simple low-pass filtering) of the designed FIR filter. Modifications are proposed to twicing and sharpening structures to regain the linear phase characteristics that are lost in conventional twicing and sharpening operations. The efficacy of all these transformed filters in minimising EMG artefacts is studied, using SNR improvements as a performance measure for simulated signals. Time plots of the signals are also compared. Studies show that the modified sharpening structure is superior in performance to all other proposed methods. These algorithms have also been applied to real or recorded EMG-contaminated EEG signal. Comparison of time plots, and also the output SNR, show that the proposed modified sharpened structure works better in minimising EMG artefacts compared with other methods considered.     

Spatial filters based on independent component analysis for magnetic noise reduction in the magnetocardiogram

A spatial filter design method to reduce magnetic noise in the magnetocardiogram (MCG) is introduced. Based on the facts that external magnetic noise appearing on multichannel MCG sensors is independent of the cardiac signals and that there is strong spatial correlation among the channels, the independent component analysis (ICA) method was applied to extract the noise components from the measured MCG signals. After extraction of the noise components in a given time period using ICA, a spatial filter was made to reduce the noise components in subsequently acquired MCG signals. In experimental studies of nine healthy volunteers, the spatial filters improved the signal-to-noise ratio of the MCG signals by about 500% on average. This spatial filtering method can be used for measurements of MCG signals in a magnetically noisy environment.     

Adaptive filtering in biological signal processing

The high dependence of conventional optimal filtering methods on the a priori knowledge of the signal and noise statistics render them ineffective in dealing with signals whose statistics cannot be predetermined accurately. Adaptive filtering methods offer a better alternative, since the a priori knowledge of statistics is less critical, real time processing is possible, and the computations are less expensive for this approach. Adaptive filtering methods compute the filter coefficients "on-line", converging to the optimal values in the least-mean square (LMS) error sense. Adaptive filtering is therefore apt for dealing with the "unknown" statistics situation and has been applied extensively in areas like communication, speech, radar, sonar, seismology, and biological signal processing and analysis for channel equalization, interference and echo canceling, line enhancement, signal detection, system identification, spectral analysis, beamforming, modeling, control, etc. In this review article adaptive filtering in the context of biological signals is reviewed. An intuitive approach to the underlying theory of adaptive filters and its applicability are presented. Applications of the principles in biological signal processing are discussed in a manner that brings out the key ideas involved. Current and potential future directions in adaptive biological signal processing are also discussed.     

Image filtering for improved dose resolution in CT polymer gel dosimetry

X-ray computed tomography (CT) has been established as a feasible method of performing dosimetry using polyacrylamide gels (PAGs). A small density change occurs in PAG upon irradiation that provides contrast in PAG CT images. However, low dose resolution limits the clinical usefulness of the technique. This work investigates the potential of using image filtering techniques on PAG CT images in order to reduce image noise and improve dose resolution. CT image noise for the scanner and protocol used for the gel images is analyzed and found to be Gaussian distributed and independent of the contrast level in the images. As a result, several filters for reducing spatially invariant noise are investigated: mean, median, midpoint, adaptive mean, alpha-trimmed mean, sigma mean, and a relatively new filter called SUSAN (smallest univalue segment assimilating nucleus). All filters are applied, using 3x3, 5x5, and 7x7 pixel masks, to a CT image of a PAG irradiated with a stereotactic radiosurgery dose distribution. The dose resolution within 95% confidence (D(delta)95%) is calculated and compared for each filtered image, as well the unfiltered image. In addition, the ability of the filters to maintain the spatial integrity of the dose distribution is evaluated and compared. Results clearly indicate that the filters are not equal in their ability to improve D(delta)95% or in their effect on the spatial integrity of the dose distribution. In general, increasing mask size improves D(delta)95% but simultaneously degrades spatial dose information. The mean filter provides the greatest improvement in D(delta)95%, but also the greatest loss of spatial dose information. The SUSAN, mean adaptive, and alpha-trimmed mean filters all provide comparable, but slightly poorer dose resolution. In addition, the SUSAN and adaptive filters both excel at maintaining the spatial distribution of dose and overall are the best performing filters for this application. The midpoint filter, normally useful for Gaussian noise, is poor all-round, dramatically distorting the dose distribution for masks greater than 3x3. The median filter, a common edge preserving noise reduction filter, performs moderately well, but artificially increases high dose gradients. The sigma filter preserves the spatial distribution of dose very well but is least effective at improving dose resolution. In summary, dose resolution can be significantly improved in CT PAG dosimetry through postprocessing of CT images using spatial noise reduction filters. However, such filters are not equal in their ability to improve dose resolution or to maintain the spatial integrity of the dose distribution and an appropriate filter must be chosen depending on clinical demands of the application.     

MUAP Number Estimates in Surface EMG: Template-Matching Methods and Their Performance Boundaries

Estimates of the number of motor unit action potential (MUAP)s appearing in the surface electromyogram (EMG) signal, which offers potentially valuable information about motor unit recruitment and firing rates, are likely to provide a more accurate reflection of the neural command to muscle than are current EMG quantification methods. In this paper, we show that the basic shapes of surface MUAPs recorded from the first dorsal interosseous (FDI) muscle can ideally be represented by a small number of waveforms. On the basis of this, we seek to estimate the number of MUAPs present in standard surface EMG records, using template-matching techniques to identify MUAP occurrences. Our simulation study indicates that the performance of template-matching methods for MUAP number estimation is mainly constrained by the MUAP superposition in the signal, and the maximum number of MUAPs allowed in the signal for a good estimation is determined by the duration of MUAPs. To further explore this from experimental surface EMG signals, we compare the recordings from a selective multiple concentric ring electrode against those derived from a standard differential EMG electrode situated over the same muscle. We conclude that the ring surface electrode only slightly reduces the MUAP duration and the less MUAP superposition rate contained in the signal is mainly achieved by reducing the pick up area of the electrode. Using a template-matching method, although the number of MUAPs can be approximately estimated based on a very selective surface EMG recording at low force levels, the maximum number of MUAPs correctly estimated from the surface EMG is constrained by the MUAP duration.     

Reduction of noise from magnetoencephalography data

A noise reduction method for magnetoencephalography (MEG) data is proposed. The method is a combination of Kalman filtering and factor analysis. A statespace model for a Kalman filter was constructed using the forward problem in MEG measurement. Factor analysis provide estimations of noise covariances required by the Kalman filter to eliminate independent additive sensor noise. The proposed method supports independent component analysis (ICA), which is difficult to use in MEG analysis owing to the sensor noise. Numerical experiments were conducted to investigate the performance of the proposed method. In a single dipole case where the maximum signal-to-noise ratio (SNR) was — 10 dB, approximately equivalent to raw MEG data, noise-free signals were successfully estimated from noisy data; a 0.02 s delay of the peak latency and 15–40% of attenuation of the peak amplitude were observed. Moreover, in a multiple dipole case, independent components preprocessed with the proposed method had high correlation, 0.88 at the lowest, with correlation of 0.69 and 0.52 for those preprocessed with conventional bandpass filters. The results show that the noise reduction method reduces sensor noise effectively. High SNR-independent components are obtained by the proposed method. Real MEG data analysis was also demonstrated. The proposed method extracted auditory evoked responses from unaveraged single-trial data.     

Noise reduction in biological step signals: application to saccadic EOG

A weighted filter for noise reduction in nonrecurrent step signals where adaptive filtering cannot be applied is described. An optimal correction of a conventional finite impulse response (FIR) filter is achieved by using a priori knowledge of noise variance and a continuous estimation of the error signal's power. The weighted filter provides an optimal compromise between noise filtering and distortionless tracking. The prior knowledge required is that of the noise power and the lowest frequency in the noise spectrum. Application of the weighted filter to the saccadic electro-oculogram (EOG) results in better estimations of saccade duration and velocity.     

超声多普勒血流和管壁搏动信号的分离研究

目的针对超声多普勒血流检测中,传统的高通滤波法在滤除管壁搏动信号的同时也会滤除低频血流信号的问题,本研究提出一种以心电信号（electrocardiography,ECG）作为参考信号的自适应滤波的方法消除管壁干扰。方法包括两方面：其一,采用心电信号作为参考信号对超声多普勒信号进行自适应滤波;其二,采用多级自适应滤波并选择不同的参考信号的滤波方案。分别使用上述方法和高通滤波法对仿真的超声多普勒信号进行处理,并将结果进行比较。结果与传统的高通滤波法相比,该方法在有效抑制管壁搏动信号的同时保留一部分低频血流信号成分。结论该方法能较准确地提取出完整的血流超声多普勒信号,具有一定的临床应用价值。     

Transient noise characterization and filtration in CCD cameras exposed to stray radiation from a medical linear accelerator

Charge coupled devices (CCDs) are being increasingly used in radiation therapy for dosimetric purposes. However, CCDs are sensitive to stray radiation. This effect induces transient noise. Radiation-induced noise strongly alters the image and therefore limits its quantitative analysis. The purpose of this work is to characterize the radiation-induced noise and to develop filtration algorithms to restore image quality. Two models of CCD were used for measurements close to a medical linac. The structure of the transient noise was first characterized. Then, four methods of noise filtration were compared: median filtering of a time series of identical images, uniform median filtering of single images, an adaptive filter with switching mechanism, and a modified version of the adaptive switch filter. The intensity distribution of noisy pixels was similar in both cameras. However, the spatial distribution of the noise was different: The average noise cluster size was 1.2 +/- 0.6 and 3.2 +/- 2.7 pixels for the U2000 and the Luca, respectively. The median of a time series of images resulted in the best filtration and minimal image distortion. For applications where time series is impractical, the adaptive switch filter must be used to reduce image distortion. Our modified version of the switch filter can be used in order to handle nonisolated groups of noisy pixels.     

Single motor unit analysis from spatially filtered surface electromyogram signals. Part 2: Conduction velocity estimation

The aim of the study was to compare experimentally conduction velocity (CV) estimates obtained with different estimation methods based on surface electromyogram (EMG) signals detected using five spatial filters. The filters investigated were the longitudinal single and double differential, transverse single and double differential, and normal double differential. The same surface EMG signals detected as described in Part 1 were used in this work. CV was estimated with four commonly used delay estimation techniques, i.e. from the distance between the peak values of two waveforms (with and without polynomial interpolation around the peak), and by the maximum likelihood estimate (MLE) based on two or more surface EMG channels. The average standard deviation of CV estimation (for all the MUs and the two muscles together) was 0.61 ms⁻¹ and 0.79 ms⁻¹ for the peak method, with and without interpolation, respectively, and 0.50ms⁻¹ and 0.31 ms⁻¹ for the MLE method, from two and more surface EMG channels, respectively. Moreover, the mean of CV estimates varied by as much as 1 ms⁻¹ depending on the spatial filter used and the method adopted for CV estimation. Considering the dependence on the spatial filter only, the average (over all estimation methods) CV estimates obtained with the five spatial filters were 4.32 ms⁻¹ (normal double differential), 4.23ms⁻¹ (longitudinal double differential), 4.61 ms⁻¹ (transverse double differential), 4.64ms⁻¹ (transverse single differential) and 4.03 ms⁻¹ (longitudinal single differential). It was concluded that the comparison of single MU CV values obtained in different studies is critical if different spatial filters and processing techniques are used for their estimation. Higher estimates of CV were attributed to a smaller reduction in non-travelling signal components and thus were assumed to be positively biased.     

Multi-adaptive filtering technique for surface somatosensory evoked potentials processing

Somatosensory evoked potential (SEP) testing has been widely applied to diagnosis of various neurological disorders. However, SEP recorded using surface electrodes is buried in noises, which makes the signal-to-noise ratio (SNR) very poor. Conventional averaging method usually requires up to thousands of raw SEP input trials to increase the SNR so that an identifiable waveform can be produced for latency and amplitude measurement. In this study, a multi-adaptive filtering (MAF) technique, emerging from the combination of well-developed adaptive noise canceller and adaptive signal enhancer, is introduced for fast and accurate surface SEP extraction. The MAF technique first processes the raw surface recorded SEP by the Canceller with a reference noise channel of background noise for adaptive subtraction before entering the Enhancer. The MAF was verified by filtering simulated SEP signals in which electroencephalography and Gaussian noise of different SNRs were added. It was found that the MAF could effectively suppress the noise and enhance the SEP components such that the SNR of the SEP is improved. Results showed that MAF with 50 input trials could provide similar performance in SEP detection to those extracted by the conventional averaging method with 1000 trials even at an SNR of -20 dB.     

一种基于自适应的新滤波技术

在心电信号的采集过程中,不可避免地会混入肌电噪声和各种干扰信号,为节获得含有较小噪声的ＥＣＧ信号,便于分析,需要对采集到的ＥＣＧ信号作消噪处理。     

Median averaging of electromyographic motor unit action potentials: comparison with other techniques

A technique of extracting individual motor unit action potentials (MUAPs) from EMG signals by median averaging, a modification of an existing method, is presented. To compare different techniques of MUAP extraction, 89 MUAPs were recorded with a concentric needle electrode in the brachial biceps muscle of normal subjects and patients with nerve and muscle diseases. MUAPs were also extracted by another method, called split-sweep median averaging, in which alternate MUAP discharges are averaged independently in two computer buffers until the two averaged signals appear equal on visual inspection by the operator. The amplitude, area, area: amplitude ratio, duration and number of phases and turns of each extracted MUAP were determined by each technique. Overall, there was a strong correlation between all features of the MUAPs extracted by median and splitsweep averaging, although the latter method required, on average, twice as many MUAP discharges to produce acceptable signals. We thus conclude that median averaging is a fast and accurate method that requires relatively few MU discharges to extract MUAP signals from spurious background signals.