Eric Laciar

Artifact removal from EEG signals using adaptive filters in cascade

Artifacts in EEG (electroencephalogram) records are caused by various factors, like line interference, EOG (electro-oculogram) and ECG (electrocardiogram). These noise sources increase the difficulty in analyzing the EEG and to obtaining clinical information. For this reason, it is necessary to design specific filters to decrease such artifacts in EEG records. In this paper, a cascade of three adaptive filters based on a least mean squares (LMS) algorithm is proposed. The first one eliminates line interference, the second adaptive filter removes the ECG artifacts and the last one cancels EOG spikes. Each stage uses a finite impulse response (FIR) filter, which adjusts its coefficients to produce an output similar to the artifacts present in the EEG. The proposed cascade adaptive filter was tested in five real EEG records acquired in polysomnographic studies. In all cases, line-frequency, ECG and EOG artifacts were attenuated. It is concluded that the proposed filter reduces the common artifacts present in EEG signals without removing significant information embedded in these records.

Commanding a robotic wheelchair with a high-frequency steady-state visual evoked potential based brain–computer interface

This work presents a brain-computer interface (BCI) used to operate a robotic wheelchair. The experiments were performed on 15 subjects (13 of them healthy). The BCI is based on steady-state visual-evoked potentials (SSVEP) and the stimuli flickering are performed at high frequency (37, 38, 39 and 40 Hz). This high frequency stimulation scheme can reduce or even eliminate visual fatigue, allowing the user to achieve a stable performance for long term BCI operation. The BCI system uses power-spectral density analysis associated to three bipolar electroencephalographic channels. As the results show, 2 subjects were reported as SSVEP-BCI illiterates (not able to use the BCI), and, consequently, 13 subjects (12 of them healthy) could navigate the wheelchair in a room with obstacles arranged in four distinct configurations. Volunteers expressed neither discomfort nor fatigue due to flickering stimulation. A transmission rate of up to 72.5 bits/min was obtained, with an average of 44.6 bits/min in four trials. These results show that people could effectively navigate a robotic wheelchair using a SSVEP-based BCI with high frequency flickering stimulation.

Automatic detection of drowsiness in EEG records based on multimodal analysis

Drowsiness is one of the main causal factors in many traffic accidents due to the clear decline in the attention and recognition of danger drivers, diminishing vehicle-handling abilities. The aim of this research is to develop an automatic method to detect the drowsiness stage in EEG records using time, spectral and wavelet analysis. A total of 19 features were computed from only one EEG channel to differentiate the alertness and drowsiness stages. After a selection process based on lambda of Wilks criterion, 7 parameters were chosen to feed a Neural Network classifier. Eighteen EEG records were analyzed. The method gets 87.4% and 83.6% of alertness and drowsiness correct detections rates, respectively. The results obtained indicate that the parameters can differentiate both stages. The features are easy to calculate and can be obtained in real time. Those variables could be used in an automatic drowsiness detection system in vehicles, thereby decreasing the rate of accidents caused by sleepiness of the driver.

Improved alignment method for noisy high-resolution ECG and Holter records using multiscale cross-correlation

The coherent signal averaging process requires accurate estimation of a fiducial point in all beats to be averaged. The temporal cross-correlation between each detected beat and a template beat is the typical alignment method used with high-resolution electrocardiogram (HRECG) records. However, this technique does not produce a precise fiducial mark in records with high noise levels, like those found in Holter HRECG systems. In this study, we propose a new alignment method based on the multiscale cross-correlation between the template and each detected beat. We report the results of tests comparing multiscale and temporal methods for 3000 beats of simulated HRECG records corrupted separately with white noise, electromyographic noise and power line interference (50 Hz) of different root mean square levels. A second study with simulated records constructed from real Holter HRECG records is also presented. The results indicate that the multiscale alignment method produces a lower trigger jitter than the temporal method in all tests. We conclude that the proposed alignment method can be used in HRECG records with high noise levels.

An epileptic seizures detection algorithm based on the empirical mode decomposition of EEG

Epilepsy is a neurological disorder that affects around 50 million people worldwide. The seizure detection is an important component in the diagnosis of epilepsy. In this study, the Empirical Mode Decomposition (EMD) method was proposed on the development of an automatic epileptic seizure detection algorithm. The algorithm first computes the Intrinsic Mode Functions (IMFs) of EEG records, then calculates the energy of each IMF and performs the detection based on an energy threshold and a minimum duration decision. The algorithm was tested in 9 invasive EEG records provided and validated by the Epilepsy Center of the University Hospital of Freiburg. In 90 segments analyzed (39 with epileptic seizures) the sensitivity and specificity obtained with the method were of 56.41% and 75.86% respectively. It could be concluded that EMD is a promissory method for epileptic seizure detection in EEG records.

Novel set of vectorcardiographic parameters for the identification of ischemic patients

New signal processing techniques have enabled the use of the vectorcardiogram (VCG) for the detection of cardiac ischemia. Thus, we studied this signal during ventricular depolarization in 80 ischemic patients, before undergoing angioplasty, and 52 healthy subjects with the objective of evaluating the vectorcardiographic difference between both groups so leading to their subsequent classification. For that matter, seven QRS-loop parameters were analyzed, i.e.: (a) Maximum Vector Magnitude; (b) Volume; (c) Planar Area; (d) Maximum Distance between Centroid and Loop; (e) Angle between XY and Optimum Plane; (f) Perimeter and, (g) Area-Perimeter Ratio. For comparison, the conventional ST-Vector Magnitude (ST(VM)) was also calculated. Results indicate that several vectorcardiographic parameters show significant differences between healthy and ischemic subjects. The identification of ischemic patients via discriminant analysis using ST(VM) produced 73.2% Sensitivity (Sens) and 73.9% Specificity (Spec). In our study, the QRS-loop parameter with the best global performance was Volume, which achieved Sens=64.5% and Spec=74.6%. However, when all QRS-loop parameters and ST(VM) were combined, we obtained Sens=88.5% and Spec=92.1%. In conclusion, QRS loop parameters can be accepted as a complement to conventional ST(VM) analysis in the identification of ischemic patients.

Automatic detection of epileptic seizures in long-term EEG records

Epilepsy is a neurological disorder which affects nearly 1.5% of the world׳s total population. Trained physicians and neurologists visually scan the long-term electroencephalographic (EEG) records to identify epileptic seizures. It generally requires many hours to interpret the data. Therefore, tools for quick detection of seizures in long-term EEG records are very useful. This study proposes an algorithm to help detect seizures in long-term iEEG based on low computational costs methods using Spectral Power and Wavelet analysis. The detector was tested on 21 invasive intracranial EEG (iEEG) records. A sensitivity of 85.39% was achieved. The results indicate that the proposed method detects epileptic seizures in long-term iEEG records successfully. Moreover, the algorithm does not require long processing time due to its simplicity. This feature will allow significant time reduction of the visual inspection of iEEG records performed by the specialists.

Multiparametric detection of epileptic seizures using Empirical Mode Decomposition of EEG records

Epilepsy is a neurological disorder that affects around 50 million people worldwide. The seizure detection is an important tool for the diagnosis of epilepsy. In this study, an epileptic seizure classification method based on features of the Empirical Mode Decomposition (EMD) of EEG records is proposed. The Intrinsic Mode Functions (IMFs) of EEG records are first computed, and then several time and frequency features of IMFs are extracted. A features selection based on a Mann-Whitney test and Lambda of Wilks criterion is performed, then these parameters are used in a linear discriminant analysis (LDA) to classify epileptic seizure and normal EEG segments. The algorithm was tested in 3 intracranial channels EEG records acquired in 21 patients with refractory epilepsy and validated by the Epilepsy Center of the University Hospital of Freiburg. The signal was divided in 15 s segments. In 45517 segments analyzed (689 with epileptic seizures) the sensitivity and specificity obtained with this method were 69.4% and 69.2% respectively. It could be concluded that the developed method could be a promising tool for epileptic seizure detection in EEG records.

Rényi entropy and Lempel-Ziv complexity of mechanomyographic recordings of diaphragm muscle as indexes of respiratory effort

The study of the mechanomyographic (MMG) signals of respiratory muscles is a promising technique in order to evaluate the respiratory muscles effort. A new approach for quantifying the relationship between respiratory MMG signals and respiratory effort is presented by analyzing the spatio-temporal patterns in the MMG signal using two non-linear methods: Rényi entropy and Lempel-Ziv (LZ) complexity analysis. Both methods are well suited to the analysis of non-stationary biomedical signals of short length. In this study, MMG signals of the diaphragm muscle acquired by means of a capacitive accelerometer applied on the costal wall were analyzed. The method was tested on an animal model (dogs), and the diaphragmatic MMG signal was recorded continuously while two non anesthetized mongrel dogs performed a spontaneous ventilation protocol with an incremental inspiratory load. The performance in discriminating high and low respiratory effort levels with these two methods was analyzed with the evaluation of the Pearson correlation coefficient between the MMG parameters and respiratory effort parameters extracted from the inspiratory pressure signal. The results obtained show an increase of the MMG signal Rényi entropy and LZ complexity values with the increase of the respiratory effort. Compared with other parameters analyzed in previous works, both Rényi entropy and LZ complexity indexes demonstrates better performance in all the signals analyzed. Our results suggest that these non-linear techniques are useful to detect and quantify changes in the respiratory effort by analyzing MMG respiratory signals.

Sleep apnea detection based on spectral analysis of three ECG - derived respiratory signals

An apnea detection method based on spectral analysis was used to assess the performance of three ECG derived respiratory (EDR) signals. They were obtained on R wave area (EDR1), heart rate variability (EDR2) and R peak amplitude (EDR3) of ECG record in 8 patients with sleep apnea syndrome. The mean, central, peak and first quartile frequencies were computed from the spectrum every 1 min for each EDR. For each frequency parameter a threshold-based decision was carried out on every 1 min segment of the three EDR, classifying it as ‘apnea’ when its frequency value was below a determined threshold or as ‘not apnea’ in other cases. Results indicated that EDR1, based on R wave area has better performance in detecting apnea episodes with values of specificity (Sp) and sensitivity (Se) near 90%; EDR2 showed similar Sp but lower Se (78%); whereas EDR3 based on R peak amplitude did not detect appropriately the apneas episodes reaching Sp and Se values near 60%.