The development and refinement of instrumentation has been a great asset in the diagnosis of neurologic diseases. With advances in instrumentation, however, physiologists are in danger of making more technologically advanced misinterpretations than previously, so it is important to have an understanding of the basic functions and limitations of modern instrumentation. This chapter will enhance the practitioner’s ability to understand how the instrumentation and its limitations may influence the interpretation of signals recorded during routine electrophysiologic studies.
Major components of an electrodiagnostic instrument
The major components of an electrodiagnostic instrument are shown in Figure 2-1 .
The electrode is the interface between the patient and the instrumentation. The proper application and use of electrodes is one of the most fundamental requirements for obtaining good signals, but it is often neglected. Electrode characteristics can affect the response. Electrodes can be classified into at least two different types: (1) surface and (2) needle.
Surface electrodes applied with conductive gel form a battery. The voltage of the battery (usually less than 600 mV) depends primarily on the type of electrode material and secondarily on how good a contact is achieved at the microscopic level. The electrode metals are usually not homogeneous and consist of numerous microscopic or sometimes visible grains. Each grain produces a slightly different battery voltage. The electrolyte is assumed to be the gel, but sweat and serum change the concentrations of sodium and other ions, thus affecting the impedance and voltage. (This is the basis of the galvanic skin response, or GSR.) Most of the battery effect on the amplifier’s active input is canceled by an equivalent battery on the reference input. Direct current blocking in the amplifier removes any imbalance. If the electrodes move or the patient sweats, however, the small changes in potential can easily be larger than the signals of interest. Abrading the skin with a ground quartz suspension (OMNIPREP) or puncturing the skin completely eliminates the GSR and much of the movement artifact.
Increasing the homogeneity of the material results in quieter electrodes. Silver electrodes can be corroded in a controlled manner to form a uniform silver chloride finish, which has noise and impedance characteristics better than those of the bare metal. If a silver electrode is abraded, its performance is reduced, and rechloriding or disposal should be considered. Aluminum electrodes form an aluminum oxide layer, which is very uniform and very quiet but also has a very high resistance. Aluminum electrodes are almost purely capacitive and do not pass low frequencies effectively. Tin, platinum, stainless steel, gold, and carbon are also used as fairly stable materials for electrodes.
Needle electrodes pose other problems. Microscopic burrs on the leading edge of the needle damage muscle as it penetrates, giving false evidence of injury in electromyography. These burrs can be detected by passing the needle through a cotton gauze. Monopolar needles have a thin layer of Teflon (polytetrafluoroethylene) or parylene insulating all but the tip. If the insulation cracks or abrades, or if there is a break in the insulation, the needle will be noisy and should be discarded. Materials, manufacturing problems, and damage during transport and handling can also result in increased noise, even in the absence of visible defects. Additional information on the care and testing of needles and electrodes is found elsewhere.
Electrode materials should not be mixed. The battery potentials created by two different materials will not cancel, and if direct-current (DC) blocking does not occur at the first stage of the amplifier, a large offset will be present. This offset may saturate the amplifier or decrease the headroom for saturation, so that clipping of the waveform occurs. The offset may contribute to unacceptable shock artifact. It may also change the operating point of the amplifier, which will degrade noise and performance. Fortunately, most modern amplifiers are designed to tolerate electrode offset.
The amplifier increases the amplitude of the desired response while it rejects unwanted noise. The first and most crucial stage of the amplifier consists of a differential input. A differential input amplifies the difference in potential presented at its two inputs (active and reference), while rejecting any signal common to both of these inputs. Reference is often used to mean a neutral input, but this only refers to a location on the body that has very little signal compared with the site of the active electrode. Both active and reference inputs are equally effective at generating potentials, and there is no neutral input on a differential amplifier.
An amplifier’s ability to reject common signals is known as its common mode rejection ratio (CMRR). The higher the CMRR, the better the rejection. Another important parameter of the amplifier input is the input impedance. Input impedance has resistive, capacitive, and inductive components. An input impedance of 10 Mohm or higher is desirable because a low input impedance attenuates the signal slightly and degrades the active-to-reference signal matching necessary for high CMRR. The higher-frequency components of a response are affected more by the input capacitance than by the input resistance. Input inductance is usually negligible. Another amplifier differential input characteristic of concern is input voltage noise and input current noise generated by the input circuitry itself. Input noise is added to the response signal.
Gain and Sensitivity
Amplifier gain describes how much the input signal is increased in voltage. The units are volts per volt, and gains of 10 to 10,000 are common. Display sensitivity describes the visible waveform and is expressed as volts per division or volts per centimeter. Smaller numerical values represent increased sensitivity; thus, 1 mV/cm is more sensitive than 10 mV/cm. A graphic display shows a vertical deflection proportional to the voltage, and changing the gain alters the size of the display. A computer displays the digital representation of the analog signals, which maintains the concept of sensitivity at any convenient gain setting for which the amplifier is designed. In most newer systems, the amplifier gain either is fixed or has a few discrete steps, and the display system is changed digitally.
The stages following the differential input amplify and filter the response signal. Low-cut (low-frequency cutoff) and high-cut (high-frequency cutoff) filters are used to narrow the frequency range of the incoming signal ( Fig. 2-2 ), and thus eliminate that portion of the noise outside the bandpass of the response signal. (Signal processing textbooks generally refer to highpass, lowpass, and bandpass filters for mathematical reasons. A bandpass from 10 to 1,000 Hz has a 10-Hz highpass and a 1,000-Hz lowpass. The designations low-cut and high-cut sidestep this confusing terminology and are used in this chapter.) The signal will also be affected if it has frequency components outside the bandpass; the filters are therefore adjustable to keep most of the signal and reject most of the noise. A component of the noise will always overlap the signal and cannot be reduced without distorting the signal ( Fig. 2-3 ).
Notch filters provide precise band-reject capability and are tuned for 50- or 60-Hz operation. The “Q” (quality) of a filter is a mathematical measure of its resonance. High-Q filters respond to a narrow but precise range of frequencies. They are used for 50- and 60-Hz notches because signals that are only a few Hertz above or below the notch frequency are passed transparently. Low-Q filters respond to a wide range of frequencies and are used for bandpass applications. Comb filters function like multiple-notch filters, which are tuned to successive harmonics of the mains frequency ( Fig. 2-4 ). Combs remove these harmonics, which make the “buzzing” sound commonly heard from 50- or 60-Hz interference.
The cutoff frequency of a filter is the frequency at which the output power is half of the input power (–6 dB) or the output voltage is 0.707 times the input voltage (–3 dB). Except for brickwall filters, the output power increases or decreases smoothly with frequency, as shown in Figure 2-2 . Changes in frequency are measured in octaves (doubling or halving of the frequency) or decades (increases or decreases tenfold). The simplest filters have a single pole and will roll off, or attenuate, the signal by 6 dB for every doubling of the frequency (6 dB per octave or 10 dB per decade), but they will also cause attenuation and phase shift well away from the −3 dB point, as shown in Table 2-1 . (The word pole is an engineering term used in describing transfer functions; one pole represents a single resistance-capacitance [RC] filter.)
|Input (Hz)||Amplitude||Percent Decrease||dB Decrease|
To separate noise and signal more effectively, steeper filters are used. Two-pole and four-pole filters are common. Each additional pole adds 6 dB per octave (10 dB per decade) attenuation. These filters are usually more than cascaded one-pole filters, having feedback and feedforward paths. Varying the amount of feedback and feedforward varies the overshoot, phase, rolloff characteristics, and amplitude ringing. Special cases of filters have specific designations, but all are part of a continuum consisting of just a few distinct topologies ( Fig. 2-5 ). Thus, the Butterworth has the flattest passband at the expense of poor rolloff; the Bessel has constant phase delay for all frequencies; the Chebyshev has maximal transition steepness at the expense of passband ripple; the elliptic filter has infinite rolloff with rebound in the stopband; the notch is a special elliptic filter that rebounds back to 0 dB; and a brickwall filter has infinite slope without rebound at the expense of maximum “ringing.” (For analog circuits, anything over 100 dB per decade is a brickwall.) Further discussion of analog filters is provided elsewhere.
Analog-to-digital conversion requires a circuit to “freeze” the signal for a few microseconds, the sample and hold (S/H), and a circuit to convert the amplitude of the “frozen” signal to a digital value, the analog-to-digital converter (ADC). Sometimes these circuits are implemented in one device, and so they will be referred to collectively as the ADC. The ADC is specified by its conversion rate and its resolution (number of bits).
The digital section consists of three major parts: processor, memory, and averager. The processor is the “brain” of the instrument; it coordinates all data flow and interface functions. Processors are classified by the number of bits processed in parallel and by the processing speed. Memory is used for processor instruction storage and for digitized signal storage. The amount of memory is expressed in bytes. The averager adds and scales synchronized signal responses to improve the signal-to-noise ratio and may be implemented by the processor.
Advantages of Digital Circuitry
Operations on digital signals are precise. Adding two analog signals gives a result with a percentage error, and the errors are cumulative. Two digital signals added together will always give precisely the same result. The chip or the components that add two analog signals will shift or drift with time, temperature, humidity, power supply voltage, and other factors. They may require calibration or compensation, and sometimes they cannot be made to work at all. The chip that adds two digital signals always gives exactly the same answer and is insensitive to its environment. Digital systems eliminate analog drift with time and temperature, and eliminate some of the need to recalibrate. Analog component values vary, but digital coefficients are absolute. A capacitor with ± 0.1 percent tolerance in a filter is an exotic part, but a 16-bit digital system has 0.001 percent accuracy, is easy and inexpensive to build, and never changes with time or temperature. Every digital unit is also exactly like every other unit, so fabrication and characterization are simplified. Digital systems can perform functions that are not practical and sometimes not possible with analog systems. Almost all analysis is easier to perform digitally. Processor performance has increased tenfold with each of the last two editions of this book, and the price and power requirements have fallen. There are now few problems or solutions that are not easier, less expensive, and more reliable to solve or implement digitally.
In most electrodiagnostic instruments, analog filters are used sparingly and have been replaced by digital filters. Digital filters can duplicate analog filters, but they can also create classes of filters not readily implemented from analog components. If implemented correctly, multipole digital filters can be superior to the analog equivalent. Multipole analog filters require components that are subject to temperature and aging variations; these act to “detune” the filter.
Two major types of digital filters are the infinite impulse response (IIR) and the finite impulse response (FIR) filters. In an IIR filter ( Fig. 2-6 ), a portion of the output data is fed back to the input. If the output does not feed back, the filter is nonrecursive and is classified as an FIR filter. The feedback term in an IIR represents the contribution of previous data points to the output. Because only a few terms are needed (two terms for each two poles), efficient filters are realizable. IIRs act much like analog filters. The infinite means that, like an RC network, the output approaches its final value asymptotically.
FIRs ( Fig. 2-7 ) compute each output from weighted portions of a limited number of past, present, and future input data points. Each point used in the computation is called a filter tap, and each tap requires a multiply-and-accumulate operation. If the FIR is symmetric (i.e., uses the same number of taps and matching coefficients into the future as into the past), then there is zero phase shift. Evoked potentials can be smoothed without changing their latency. (Three-point smoothing algorithms are FIRs.) Obviously, the future is never known in the real world, even for the upcoming few milliseconds. An FIR filter avoids this problem by delaying the output for half the number of taps and moving the time reference by the same amount. In a real-time system, all frequencies are delayed by the same amount and only the relative phases have zero shift. This trick has a small price. Steep-skirted filters are accompanied by ringing whenever an edge or impulse occurs; this is called Gibbs’ phenomenon . Because zero has been shifted out in time, half of the ringing occurs before the impulse and can sometimes be seen before the stimulus. The additional peaks created by Gibbs’ phenomenon are artifact, as is their occurrence before the stimulus. Such peaks have been interpreted and presented as new responses previously hidden by the noise or as anticipatory potentials.
Digital filter characteristics, including cutoff frequency and the number of poles, are determined by their coefficients. Adjustable frequencies do not require more resistors, capacitors, and analog switches; they only require changing a set of numbers in the processor and adding computational power. High-cut, low-cut, and notch filters for multiple channels are typically implemented with a single processor.
Fast Fourier transforms (FFT) are a class of algorithms that turn time data into frequency-phase data and vice versa. They are especially convenient for looking at a signal’s frequency characteristics and for implementing brickwall or other arbitrary filters. Fast refers to algorithms implementing the Fourier transform by eliminating redundant calculations to speed up computation. A brickwall filter is implemented by performing the FFT, zeroing out all unwanted frequency terms, and then performing an inverse FFT. Brickwall filters show maximum Gibbs’ phenomenon but are maximally effective at reducing noise outside the passband.
A response can be presented visually and audibly. Historically, electromyographic instruments (EMG) used analog oscilloscope displays, where the response signal vertically deflects an electron beam as it sweeps horizontally across the face of a phosphor-coated tube. All modern systems use digital displays, which create pictures by illuminating individual pixels (dots) on a screen to create a picture. Digital displays have inherent persistence supplied by video memory instead of by long-persistence phosphors. Other advantages that oscilloscopes once had can be simulated with clever programming, and the ability to combine waveforms and graphical and textual information on one display has eliminated oscilloscope technology.
Auditory presentation of the response is useful not only in helping to classify a response but also in detecting and identifying noise. Types of interference from power lines, fluorescent lights, cathode ray tube (CRT) displays, biologic artifact such as EMG activity, electrode artifact, and sterilizing ovens are easily identified by their characteristic sounds.
Stimulation occurs when the voltage across the nerve membrane is decreased enough to initiate depolarization. The nerve has sodium ion pumps that normally maintain a resting potential, and the stimulating current must overwhelm the pumping capability. To do this requires a fairly constant charge per stimulus. The charge is the area under the curve of an amplitude–duration plot, and accounts for intensity, pulse width, and wave shape. The most common wave shape is a square wave because it is easy to generate. Other wave shapes will not change the charge requirements, although wild claims have been made to the contrary. If the charge is introduced very slowly with the use of a low-amplitude, long-duration pulse, the nerve is able to compensate partially for the stimulus and requires more total charge for depolarization. The strength–duration curve shows this relationship, which varies for different tissues.
Several caveats and exceptions are well known. If the nerve is initially hyperpolarized and then depolarized, the total charge required can be reduced. A biphasic stimulus can also be used to achieve zero net charge transfer, which may eliminate electrolysis and possible tissue injury in direct nerve stimulation. In polysynaptic systems (the brain), both inhibition and potentiation are observed with paired pulses, depending on the interstimulus intervals.
Constant , as in constant current, means that the output remains at the specified, adjustable level. Constant-current stimulators have high output impedance and allow the output voltage to change to maintain the desired current. Constant-voltage stimulators have low output impedance and allow the output current to vary to maintain the desired voltage. Other stimulators have finite output impedance and allow both voltage and current to change as the load impedance changes ( Fig. 2-8 ).
Constant current has the theoretical advantage that the product of duration and current determines the stimulator’s effectiveness. Some authors claim that constant current is less painful, but this is both subjective and sensitive to technique and methodology. Constant-current stimulators will deliver the same current as electrode gel dries out, an advantage for those who do not or cannot check electrodes.
Constant-voltage stimulators will deliver a stimulus to the nerve, even if a small amount of gel or body fluid is shorting the leads. It will charge the body capacitance quickly and then deliver full current, and so it is a more powerful stimulator.
Electrical stimulators produce large voltages that can introduce artifact into the waveform. Some of the artifact occurs if the amplifier saturates and has a long recovery time. In 1980 the authors began turning the amplifiers off during the stimulus to reduce amplifier recovery time. Clamping the stimulator output immediately after the stimulus or putting out a biphasic wave to remove the charge stored by the body, or both, is also useful. If the stimulus is completely isolated, very little current should be common to both stimulator and amplifier. Complete isolation is difficult to obtain; instead, the upgoing voltage on one stimulator electrode and the downgoing voltage on the other are designed to have equal capacitive coupling back to ground to cancel out amplifier-stimulator leakage currents.
The brainstem auditory evoked potential (BAEP) is generated by acoustic signals between 4 and 8 kHz. A 100-μsec square-wave click has most of its energy in this band. The click is amplitude-controlled between 0 and 130 dB, a range of over 3,000,000 : 1. Careful attention to noise and inadvertent feedthrough is needed for this dynamic range. Headphones capable of faithfully reproducing the electric pulse generate an auditory click. Both magnetic transducers and piezoelectric transducers are used. The magnetic transducers mounted on a headband generate a small electrically coupled artifact at the beginning of the sweep. Piezoelectric transducers are usually placed about a foot away from the ear on a hollow acoustic coupling tube, using spongy inserts to hold the tube in the ear canal and to suppress ambient noise. Because the click takes about 1 msec to traverse the tube, all components of the response are separated from the artifact by 1 msec.
Visual stimulators for eliciting evoked potentials depend on the rapidly changing contrast along the edges of checks to produce a response. Raster scan, plasma and liquid-crystal display (LCD) TV monitors and light-emitting diode (LED) checkerboards are used. The LED checkerboard generators can reverse the on-off pattern almost instantly. The raster scanned displays take from 0 to 16.7 msec to change patterns, related to the time it takes for the beam to sweep the display. This stimulus lag shows up as a smearing and an 8-msec latency shift of the responses compared with the LED responses, Other display technologies have delays that vary by manufacturer and user setup, and require calibration and normal values for the stimulator used. Check size, contrast, intensity, and the subject’s visual acuity also affect the response, as discussed in Chapter 22 .
Magnetic stimulators generate a 1- to 2-Tesla magnetic field in 50 to 100 μsec, which induces a voltage in peripheral nerves or cerebral cortex sufficient to achieve depolarization. The principle is the same as that for a transformer, in which a changing magnetic field induces a voltage around it. The stimulator is the primary of the transformer, and the cortex is the single-turn secondary. The body is almost perfectly transparent to the magnetic field, and so currents can be induced below the skin with minimal or no pain. Magnetic stimulators allow measurement of motor evoked potentials, which are discussed in Chapter 28 .
The algorithms used to control the instrument are known as the software . The resources allocated to write the software exceed the effort to design the modern electrodiagnostic instrument. To partition the design effort, most systems have multiple processors, each of which controls a portion of the system. The various stimulators and the amplifier may each have a dedicated processor and associated software or firmware. The software may reside in various formats in an instrument. Software that is programmed into nonvolatile memory is called firmware . Software resides on hard disk or CD-ROM and is loaded into system memory during initialization. The design and reliability of the software have a large influence on the utility of an instrument.
As EMG instruments, and especially the reports and data they generate, become less stand-alone and more integrated into electronic records and connected computer systems, software design becomes more demanding. The user interface becomes more critical; it must appear simple and intuitive when, in fact, increased effort is needed to make it more intuitive. The underlying software components have to be cleanly partitioned so that maintenance and testing of the hardware, acquisition, and user interface can be verified. Awareness of and adherence to software standards allow other systems to access, utilize, and display the results, and may allow collaboration in ways that proprietary solutions preclude.
Factors that reduce signal fidelity
Physiologic signals are mixed with noise. Low-level signals of all sorts are plagued with noise, and it is noise that limits the resolution and precision of the signal measurement. Noise usually refers to white noise , but several other types of noise, with multiple sources and varied solutions, are worth considering. Advances in technology have also introduced new noise sources.
Random noise or white noise sounds like a harsh “shhhhhh.” It is generated by processes that are statistical in nature, and has uniform energy in all frequency bands (energy per band = constant). A 10 k-ohm resistor at room temperature generates about 0.3 μV of white noise across its leads just lying on a bench. This thermal noise, generated by agitated electrons, places an absolute lower limit on amplifier quietness unless the system is cooled to absolute zero or unless the input impedance is reduced to zero. Passing a current through the resistor creates additional pink noise (energy per band = 1/frequency), which has a more musical “shhhhh” sound. Because cooling (of patients) is not practical, skin preparation and conductive gel are required to decrease impedance, and high-impedance amplifiers are necessary for decreasing current flow. The preamplifier has additional noise determined by engineering choices, and generally cannot be improved easily. The electroencephalogram (EEG) is nearly random noise (in evoked potential studies), as is weak background EMG activity; these are usually the dominant sources of white noise.
Impulse noise sounds like a “pop,” “crack,” or “click” and includes transistor “popcorn” noise, static discharge, EMG artifact, artifact from metal dental fillings touching intermittently, and electrode movement. Impulse noise , as used here, is present for a short time in only one epoch, unlike random noise, which is present uniformly throughout each epoch.
Mains 50- or 60-Hz interference (assumed 60 Hz in this discussion) produces a continuous audible buzz if harmonics are present, but it is inaudible or barely audible otherwise. It is induced by magnetic induction and by capacitive coupling. Harmonics are present when iron-core transformers, dimmers, and fluorescent lights are nearby. The energy is all at 60 Hz, 120 Hz, 180 Hz, and so forth, and is usually biggest in the odd harmonics (e.g., 180 Hz, 300 Hz); the energy in high-order harmonics drops rapidly.
In-Band Noise Source
Cellular telephones, high-efficiency fluorescent lights, switching power supplies for laptops, and blood pressure cuffs with digital readouts are examples of the profusion of noise sources. Regulatory mandates to control such sources are growing in response to the awareness of their adverse effect on sensitive measurements. Most of these emit electrical noise in the 1,000- to 100,000-Hz range, which either steps on the signal of interest or is poorly rejected by the amplifier. Awareness and avoidance are essential.
Synchronous noise is time-locked with stimulation and averaging. It can be generated by numerous sources:
The instrument processor generating the stimulus executes the same instruction sequence and may radiate a characteristic burst of energy.
The timer used to generate the stimulus rate may radiate electrical noise.
Electrical stimulator recovery may have an abrupt turnoff after many milliseconds.
The power supply may be modulated by the slightly increased power demands during stimulation.
Headphones with a low-frequency resonance may ring down for several milliseconds.
The patient may blink or track the target used to elicit visual evoked potentials (VEPs), thereby introducing electroretinogram (ERG) signals.
Signal-to-Noise Ratio (SNR)
The relative size of the signal to the noise determines how well the signal can be visualized or even detected. The evoked potential signal present in any one epoch is 1 to 100 times smaller than the background noise. The electrodiagnostic equipment must obtain an SNR that is better than 3 : 1 for reproducible testing. Figure 2-9 is a graphic presentation of SNR values.