Signal Processing of Biosignatures
What is auditory evoked potential?
Auditory evoked potentials test the auditory nerve and the auditory pathway in the brain stem. Electrodes placed on both ear lobes and Cz. Used in estimating hearing loss and to test on children.
What is zero padding?
- Adding zeros to a signal to get the length 2^n to be able to apply FFT. This will reduce the numerical operations for DFT. - Increasing duration to provide interpolation of a frequency between x(k) and x(k+1). This can be used to determine the location of an isolated peak. It does not improve resolution.
Electrode types EMG
- Direct contact: Ag/Au plated, reusable - Floating electrodes: in a cup with electrolyte, for dynamic movements, single use - Hydrogel electrodes: AgCl covered with sticky layer of gel, higher impedance. Can be reused/moved - Dry electrode: Ag/Au plated, reusable, pre-amplifier in electrode.
Visual display of SEMG
- Raw EMG - Processed EMG in time domain - Spectral analysis (frequency, time-frequency analysis) - Probability amplitude histogram (time domain)
Why do we use filters?
- To extract certain frequencies (bands in EEG) - Remove a specific frequency range (Notch, remove noise/artefact) - Remove high/low frequencies - Enhance certain frequencies (edge enhancement, enhancing audio signal)
4 sources of beta activity
1. Frontal beta: fairly common, may be very fast, increased with increased concentration; 2. Central beta: partly but not generally the basis of rolandic mu rhythm often mixed with mu rhythm. 3. Posterior beta: often a fast alpha equivalent, reactive like alpha rhythm to closing/opening eyes 4. Diffuse beta: no linkage to any special physiological rhythm
List 5 main areas of EEG singnal processing application.
1. Long term monitoring (monitoring anaestheisa, vigilance state or cognitive load) 2. Long term monitoring for diagnostic purposes (Epilepsy, sleep monitoring, drugs, neonatal) 3. Short term monitoring (event related potentials, injury of brain, confirming brain death, monitoring brain development) 4. Brain computer interface (Neuromodulatory therapies, gaming, communication and control) 5. Neuroscience.
Spectrogram
A 2D presentation of time-frequency dependencies of the signal
What is the difference between a digital and an analouge filter?
A digital filter is a numerical procedure that transforms or selectively extracts info from the input digital signal to an output digial signal. An analouge filter can be provided in an analytical form but also as an electrical circuit.
What is a sampled data signal?
A discrete signal with continuous amplitude. Same as digital signal.
Equiripple filters
A filter designed to provide equal attenuation in the stopband across all frequencies. Therefore we will not get the wrong impression of the magnitude of a frequency in the stopband.
What is a notch filter?
A filter that cancels a narrow band of frequencies. Can be used to cancel 50 Hz noise or the resonance of the system.
What is Welsh periodogram method?
A method to calculate PSD. The sequence is divided into smaller segments, overlapping 50%. Each segment is multiplied by a window function in time domain. FFT of each windowed segment. Results are averaged over different time windows.
Electrocardiogram
A recording of the electrical activity of the heart
What is a seizure and what is the name of the disease with frequent seizures?
A seizure is the most extreme form of synchronous brain activity and it can lead to loss of consciousness, muscles show tonic or clonic activity. Epilepsy is frequent seizures.
What is a digital signal?
A signal with discrete values on both the independent (x) and dependent (y) variable.
Force-amplitude relationship in muscles
A trained individual will have lower amplitude for the same weight. Linearity depends on the types of fibres in the muscle. Only one fbre type->linear.
Which 3 electrode types are required for EEG recording?
Active, reference (can also be active), ground. Ground electrode in an electric device is there for safety purposes and to prevent excessive noise in signal.
Explain average reference in EEG!
All active electrodes (evenly distributed over the scalp) is summed up and averaged. R=(EEG1+EEG2+...+EEG5)/5, EEG(C3)=C3-R.
Why do we use autocorrelation in EEG signal processing?
Autocorrelation is used to see the similarity between different segments of the same signal. It can be used to analyse how EEG changes over time. When autocorrelation reduces it means that signal started to change (but we cannot say in which way). Ex in EEG signal of a person with epilepsy, reduced autocorrelation means that dynamics of the signal changed e.g. seizures, or in a person with sleep problems we can monitor EEG during sleep to detect when the signal started to change due to the onset of different phases of slip, sleep spindles. Note that we cannot say what exact changes happened in the signal, autocorrelation is simply indicating the onset of some change. To be more precise about the time of changes we can use smaller time windows and calculate autocorrelation (normally up to few seconds).
Advantage of Welsh periodogram method
Averaging 'smooths' PSD so the location of peaks indicating frequencies which bear the most of energy become much clearer.
Why are both time and frequency presentation of a signal in DFT periodic?
Because we sampled our signal in both domains which caused convolution in the other domain. Convolution resulted in the periodical repetition of the signal
What is the 2 generators of synchronous EEG rhythm?
Central clock-pacemaker (the thalamus). Synaptic connection between thalamic cells for synchronisation of the rhythm of the group and thalamocortical axons pass this rhythm from thalamus to the cerebral cortex. The other is in the cerebral cortex, where cells exite and inhibit each other. (Distributed timing function)
What is the difference of coherence and correlation?
Coherence explains a similarity between two time sequences in time-frequency domain, correlation in time domain.
Isotonic contraction
Constant muscle force. Can be both concentric or eccentric.
What is period/interval analysis in EEG?
Counts the number of incidents where the EEG crosses the zero voltage line, it's not very precise. Often used to determine depth of anaesthesia. Related to the dominant frequency of EEG.
What does DFT show?
DFT presents frequency content of time series.
DIT and DIF
Decimation in time/decimation in frequency. Both require N=2^m. The idea is to sequentially decompose the original N point sequence, obtain DFT and form a weighted combination of the DFTs of either these rwo to get the DFT of the original sequence. Zero padding is used to make the length N=2^m.
What mental states are the bands connected to?
Delta 0-4 Hz, Deep sleep, infancy, brain disorders Theta 4-8 Hz Infrequent, mostly in adolescents, in adults during stress Alpha 8-13 Hz Awake or movement of body Beta: 13-30 Hz, opening/closing of eyes, concentration Gamma >30 Hz Perception and consciousness, REM sleep
EEG frequency bands
Delta 0.1-4 Hz Theta 4-8 Hz Alpha 8-13 Hz Beta 13-30 Hz Gamma >30
3 types of signals:
Deterministic: possible to define all values based on measurements in the past Random signals: cannot be predicted Stochastic signals: have a certain element of randomness but we can predict their future value based on the past values with a prediction error (biological signals)
What is the difference between DTFT and DFT?
Discrete time fourier transform is continuous function of the real variable w. Discrete fourier transform is a discrete periodic function of integer multiplies of frequency (2pi/N)
What is ECoG?
Electrocorticography is a type of electrophysiological monitoring that uses electrodes placed directly on the exposed surface of the brain to record electrical activity from the cerebral cortex
What parts are needed for EEG recording?
Electrodes, amplifier, AD converter, PC.
What is the difference between energy spectral density and power spectral density?
Energy spectral density shows how the energy of a signal is distributed over frequency. Power spectral density is obtained when the energy spectral density is divided by a finite period in which a signal was measured. PSD shows contribution of each frequency to the power of the signal. Summing up PSD over all frequencies we get total power.
Stable system
Every bounded input has a bounded output (BIBO)
Explain the differences and similarities between evoked and event related potentials!
Evoked responses are a result of processing a physical stimuli, they have short latency following the physical stimuli (typically up to 100ms); time locked to an external event (=external stimuli) Event related potentials are also the result of processing an external physical stimuli but are caused by "higher processes" that might involve memory, expectation or attention; also time locked to an event by have larger delay because higher processes. The larger latency of a response the more cognitive processing is involved. Both simple evoked responses and event related potentials require averaging in time domain of many repeated responses in order to be seen (as background EEG is not regular and get cancelled by averaging)
What are evoked responses?
Evoked responses are a result of processing a physical stimuli. They are time locked to the stimuli and repeatable. The clinical application is to check the sensory pathways. The latency time is between the stimuli and the response peak.
EMG analysis in frequency domain
FFT, power spectral density, spectrogram, coherence (2 muscles or muscle and brain)
What is the difference of FIR and IIR filters?
FIR only uses previous values of the input to create the transfer function, IIR also uses previous values of the output.
Type 2 muscle
Fast twitch, paler, lower aerobic metabolism, fatigue faster, used for power burst movements, rely on ATP for quick energy source
Energy signal
Finite energy signal with 0 average power
When do you use Welsh spectrogram to calculate PSD?
For a continuous EEG, in Matlab PSD Welsh periodogram method is used to calculate PSD (non-parametric)
What is fMRI?
Functional MRI is used to measure the blood oxygen level in different areas in the brain.
What is fNIRS?
Functional near-infrared spectroscopy is the use of near-infrared spectroscopy (NIRS) for the purpose of functional neuroimaging. Using fNIRS, brain activity is measured through hemodynamic responses associated with neuron behaviour.
What is energy spectral density?
How the energy of a signal is distributed over frequency. Only calculate when energy is finite. Summing up energy spectral densities over all frequencies gives total energy.
Discrete Wavelet Transform (DWT)
Ideally we want longer time windows for lower frequencies and shorter time windows for higher frequencies so we can catch changes of higher frequencies. This time frequency transform allows that.
Power signal
Infinite energy signal with finite average power
What is a shift-invariant system?
Input output relation is independent of the time of application of the input
Electrocardiograph
Instrument used to record the electrical activity of the heart
Sources of noise in EMG
Instrumental: - 50 Hz - Radio - Electromagnetic force from moving cables Biological: - ECG - DC current shift, moving artefact - Respiration - Cross talk from distant muscles
What is period/interval analysis in EEG?
It counts the times when EEG signal crosses 0 V-line. It is not very precise and it is used in the intraoperative monitoring of the depth of anaesthesia. Related to the dominant frequency of the EEG signal.
What is coherence?
It explains a similarity between two time sequences in time-frequency domain. (auto/cross)
What is correlation?
It is used to describe similarity between signals in time domain and it goes from 0 o 1.
What is Laplace derivation filter? (Spatial filtering Laplacian)
It's a spatial filter in time domain analysis of EEG. It enhances the contribution of EEG sources closest to the recording site and filter out contribution of distant source. 5 electrodes.
Einthoven's Law
Lead I + Lead III = Lead II at any instant in the cardiac cycle
What is MEG?
Magnetoencephalography is a functional neuroimaging technique for mapping brain activity by recording magnetic fields produced by electrical currents occurring naturally in the brain, using very sensitive magnetometers.
What is activity in EEG?
Measure of the squared standard deviation of the amplitude (variance of mean power) σa^2. Describe EEG in time domain and related to the variance of the signal. Used to describe the dynamics of the signal without analysing its frequency content.
Convolution
Measurement of the effect one signal on another signal.
Unipolar chest lead
Measures the potential variation at a point P compared to the central terminal. V1-V6 in ECG.
Types of recording in EMG
Monopolar: active on muscle, reference anywhere Bipolar: recording and reference 2 cm apart on the same muscle Multipolar: several active electrodes with the same reference
What are the three types of EEG recording/referencing?
Monopolar: all EEG electrodes have a common reference, placed where there is little brain activity, such as ear, matoid bone, tip of the nose. The common reference can also be one of the EEG electrodes. Bipolar: between 2 electrodes. It reduces common EEG activity and common noise. Average common reference: Reference is a mean value of all EEG electrodes
Isometric contraction
Muscle contracts but there is no movement, muscle stays the same length. Press hand against surface.
Concentric contraction
Muscle shortens during contraction
Eccentric contraction
Muscles lengthens during contraction. Lowering weight.
What is the 10-20 system for EEG electrodes?
The name 10-20 comes from the distance between the electrodes. The numbers increase going from the centre to left/right, odd numbers on left lobe, even numbers on the right lobe. The letter corresponds to different parts of the head/lobes. Frontal F, Temporal T, Parietal P and Occipital O, Central area C.
What features in an ideal filter is impossible to create in a real filter?
No ripples in the stop band, no overshooting in passband and a zero transition band.
What are two main types of analysis of EEG in frequency domain?
Parametric: assumes that the signal has a certain structure which can be desbribed using a small number of parameters. Not good with noisy signals. Non-parametric: does not assume that the process has any particular structure.
Describe methods for analysing EEG in time domain.
Period/interval Activity, mobility, complexity, Autoregressive modelling (parametric) Cross and autocorrelation parametric template matching, Spatial filtering-Laplacian (enhances contribution closest to the recording site) Averaging (ERP)
Explain the differences between 2 phases of sleep
REM: rapid eye movement. Low voltage, fast EEG, dreaming, body is immobilized. Non-REM: high voltage, slow EEG, energy consumption is lowered, body is capable of moving.
Discrete time Fourier transform maps time-domain sequence into the domain that is a function of the _____ variable
Real
How can we process EMG?
Rectify Smoothing Filtering 10-500 Hz Notch 50 Hz Whitening
Type 1 muscle
Red, dark, slow twitch muscle fiber which are smaller, tend to produce less overall forces, but are more energy efficient than fast twitch fiber, many mitochondria for oxygen supply. For long activities.
What is ROC?
Region of convergence, a set of points in the complex plane in which ZT summation converges
What is mobility in EEG?
The measure of the standard deviation of the slope (1st derivative of EEG) with reference to the standard deviation of the amplitude [d(Activity)/dt]/(Activity)^1/2 May be regarded as a mean frequency Describe EEG in time domain and related to the first derivative of the signal. Used to describe the dynamics of the signal without analysing its frequency content.
What is somatosensory evoked potential?
SEP happens due to electrical stimulation of a nerve. SEP is a result of a summed effect of action potentials and synaptic potentials in a volume conductor. The latency is 20 ms for the upper limb peripheral nerves. It is used to check demyelination of nerves. Pathway: afferent nerve, dorsal columns in the spinal cord, the lemniscal and thalamocortical pathways in the brain. Electrodes placed on the contralateral somatosensory area.
For which types of signal is STFT used? What assumption do we make when using STFT? How do we create STFT?
STFT is used for dynamic signals which change over time. We assume that the signal is stationary within a window in which STFT is applied. To create: Instead of applying one Window function over the whole period Ts in which we want to analyse x(t) we apply a set of smaller windows, which are typically overlapping for 50% to avoid the edge aliasing effect. In this was we can analyse how time-frequency distribution changes over time.
Name 4 types of evoked responses.
Somatosensory SSEP Brainstem auditory BSEP Visual VEP Thermal/pain EP
SEMG SENIAM (some recommendations)
Surface electromyography Surace Electromyography for Noninvasive Assessment of Muscles - electrode shape of conductive area - electrode size - inter-electrode distance - electrode material - electrode construction
What is quantisation?
Taking samples values of a dependent variable or amplitude. This determines the precision of measurement.
What is the difference between two 8-12 Hz rhythms?
The alpha rhythm occur when a person is awake, at the posterior regions of the head. The amplitude is below 50 µm and this is blocked by attention (visual). The Rolandic central mu rhythm is closely related to activity of the motor cortex. During movement of a certain body part, the amplitude drops compared to a relaxed state, at the same time mu rhythm can increase in other areas.
Wilson's central terminal
The central terminal E(WCT) is the average between the potentials in right and left arm and left leg. It is relatively constant through the cardiac cycle.
How is common noise reduced in bipolar EEG recording?
The difference of EEG activity of 2 electrodes is measured. EEG(F3C3)=EEG(F3)-EEG(C3)=(F3-A1)-(C3-A1)=F3-C3=(B(F3)+N)-(B(C3)+N)=B(F3)+B(C3) where N is noise, B is brain activity.
What is the source of the electrical activity in the EEG signal?
The flow during synaptic excitation of the dendrites of pyramidal neurons in the cerebral cortex is measured during EEG, and the signal is the result of the joint activity of thousands of neurons. The amplitude is proportional to the number of synchronously activated neurons. The EEG signal is a blurred version of reality.
What is complexity in EEG?
The number of standard slopes actually generated during the average time required to generate one standard amplitude given by Mobility. Require calculation of the second derivative Describe EEG in time domain and related to the second derivative of the signal. Used to describe the dynamics of the signal without analysing its frequency content.
Causal system
The output depends only on past and present values of the input (biological signals)
What is needed of the system to calculate ZT?
The system need to be time invariant and linear
What is the independent variable in biological signals?
Time (or space in images). The dependent variable is the amplitude.
In which three domains do we typically present transfer function of a filter?
Time domain impulse response, Laplace transform (analogue filter) and Z transform (digital filter).
What are the 4 types of EEG signal processing?
Time domain, frequency domain, time-frequency and space-time-frequency domain
When do we use autocorrelation in biological signals?
To analyse how EEG over single electrode changes over time. When auocorrelation reduces it means that signal started to change (but we cannot say in which way). For example in EEG signal of a person with epilepsy, reduced autocorrelation means that dynamics of the signal changed e.g. seizures, or in a person with sleep problems we can monitor EEG during sleep to detect when the signal started to change due to the onset of different phases of slip, sleep spindles. Note that we cannot say what exact changes happened in the signal, autocorrelation is simply indicating the onset of some change. To be more precise about the time of changes we can use smaller time windows and calculate autocorrelation (normally up to few seconds).
2 types of common EEG artefacts
Transient activity: Electrooculogram (blinking, eye movement) Muscle activity (mainly frontal and temporal) Movements of head and body Electrocardiographic activity Pulse-wave artefacts from blood pulse waves Instrumental noise: 50 Hz background noise, Low frequency drift (electrode polarisation)
What is bilinear transform?
Used to convert between Z and S transform. This is used when converting an analouge filter to a digital.
What is visually evoked potential?
Visual evoked potentials test the visual pathway, from retina to occipital cortex. Monocular stimulation, electrodes are placed on the occipital area. Prolonged P100 indicates optic nerve dysfunction
Types of muscle activity
Voluntary contractions: controlled by motor cortex Evoked contractions: reflexes or electrically induced contractions
What is an analouge signal?
When both the independent and the dependent variables are continuous.
What is power spectral density?
When energy spectral density is divided by a finite time period in which the signal was measured. It shows contribution of each frequency to the power of the signal. Summing up all gives total power.
What is a continuous signal?
When the independent variable (x) is continuous
Describe the relation between the Z transform and DFT
ZT presents a sequence x(n) as a sum of scaled, weighted sinusoid functions. DFT presents x(n) as a sum of weighted sinusoid functions (not scaled since r=1). ZT may exist even of DFT does not.
What is dB? When is it used in signal processing?
dB is a logaritmic unit that indicates the ratio of a physical quantity relative to a specified level. In signal processing we usually give info about at what frequency a signal has dropped 3 dB (halved)
