Data-recording systems

Equipment for performing visual electrodiagnostic testing of the visual system is described in this chapter. In acquiring information for this chapter, a limited survey was sent to a number of those performing electrodiagnostic testing in the clinic and laboratory to determine what equipment was being used for testing. On the basis of the responses to this survey, a broad array of equipment descriptions and sources has been collected. What was surprising was the lack of consensus about equipment for both clinical and laboratory applications. This chapter is intended as a starting resource for those who are interested in either purchasing a commercial system or designing and building their own.

A significant number of respondents to the survey indicated that their laboratories use systems that were designed and built in their own laboratories for their own specific purposes and that undergo modification as required. The need for self-built systems, it was reported, stemmed from finding that commercial systems were inadequate for the specific purposes of the laboratory or did not allow sufficient flexibility for experimental work. The cooperation between systems designers and clinical/research laboratories has led to a rapid improvement in the quality and design of commercial electrodiagnostic systems. As a result, these stand-alone systems have increasingly found their way into testing laboratories, particularly in clinical electrodiagnostic services, where they seem to dominate. However, research-oriented facilities continue to use their own systems. In acquiring a recording system, therefore, the main message that was derived from the survey seemed to be “know what you want to do with the system, and make sure that the system has enough flexibility to meet the needs of the laboratory for the expected life of the equipment.”

Components of Data Acquisition Systems

Modern recording systems for use in clinical visual electrophysiology laboratories generally follow a standard format, be they for routine clinical work or more complex research applications. They are almost inevitably built around a personal computer (Macintosh or Windows), although in some cases, this may not be apparent on casual examination.

This equipment configuration is defined largely by the problems inherent in the task.

In simple instrumentation terms, the task or objective is to record small signals in an environment that is far from ideal. These signals generated from the various structures of the visual system fall within the range of 10nV to 1000µV in terms of amplitude and 0.1–300Hz for frequency. As a signal generator, the patient is not ideal, having a low signal-to-noise ratio and no direct connection points to the system under investigation. The recording environment is also noisy electrically and becomes increasingly degraded with the proliferation of electronic equipment. Instrumentation systems are designed in part to circumvent these problems.

The main stages of a typical system are described (figure 16.1), along with some of the specific problems associated with each stage.

Figure 16.1.  

System overview.

System Overview

The first component is the stimulator, commonly either a Ganzfeld bowl or a pattern stimulator, which provides a controlled light source that may be synchronized with the data acquisition components. These units are dealt with in more detail elsewhere.

Regardless of which stimulus source is used, it is essential that it is calibrated correctly and that the calibration is repeated at suitable intervals. If this is not done, there can be no certainty that variations from the norm are due to the patient and not to changes in stimulus parameters. Most visual stimulators (including monitors, simple xenon flashes, or LCD screens) cannot be assumed to be stable. All responses are sensitive to variations in stimulus parameters, many in a nonlinear fashion. Therefore, without precise definition of the stimulus, uncertainty relating to the responses that are produced is inevitable. For details of stimulus calibration, refer to the ISCEV guidelines.¹

The Patient

In some tests, the patient must remain immobile in the same position for extended intervals, and care should be taken to ensure that the patient is not only correctly positioned for delivery of the stimulus, but also in a comfortable position that can be maintained for the duration of the examination. Incorrect positioning of the patient will mean that the stimulus varies between patients and with time in the same patient. In addition, the quality of the recorded signals will deteriorate. A patient who is comfortable, relaxed, and well supported will generate much lower levels of artifact.

Electrodes

The electrodes through which signals are recorded are, superficially at least, the simplest piece of equipment in the chain. They generally consist of a piece of metal or other conductive material attached to a length of wire with a plug on the other end. This piece of equipment, frequently dismissed so briefly, is the most critical link in the chain after the patient. Any lack of care in the maintenance or attachment of the electrodes will render any data recorded at best suspect and at worst misleading and is by far the most common cause of obtaining poor results.

Although the complex electrochemistry involved in the electrode-patient interface is beyond the scope of this article, whichever of the numerous types of electrodes used for visual electrophysiology are selected for a specific application, the importance of rigorous standards of care in the application and maintenance of electrodes cannot be overstressed. The importance of good electrode application technique cannot be overstressed. If this stage of preparation is not carried out to the highest possible technical standards, then the best equipment available will not give good reliable results. The same can be said of the care and maintenance of electrodes. (See chapter 17.)

Amplifiers

An amplifier is simply a device for increasing the level of a signal (figure 16.2). To cover the range of visual evoked potentials that are normally investigated, a gain ranging from 1000 to 100,000 times is required. Electrophysiology amplifiers have a high input impedance and low output impedance. Although the input impedance is high, it is not infinite, and therefore, it will draw current from the signal source. The amplifiers used in visual electrophy-siological applications are differential or balanced input devices. These differ from the more common type of amplifier in that they have two signal inputs rather than one. The first is the active input, while the second is the reference, and the voltage difference between these two is amplified.

Figure 16.2.  

Single-ended and diferential amplifiers.

In the single-ended amplifier, the signal between ground and the input is amplified; this is adequate in an environment in which the noise signal is of such a low level that it may be neglected. In a biological setting, the noise signal, that is, any signal other than that which one wishes to record, is one or more orders of magnitude greater than the signal to be recorded, so the single-ended amplifier is of little practical use.

The differential amplifier amplifies the difference in signals between its two inputs. Any common mode signal, that is, a signal occurring equally at both inputs, is rejected. This common mode signal will, for example, by virtue of the relative distance of its source from the recording site, appear with identical amplitudes at both inputs to the amplifier and therefore will not be amplified, as would the true signal. The ability of an amplifier to reject common mode signals is known as the common mode rejection ratio (CMRR). Modern amplifiers will have a CMRR of >100dB. However, this is for idealized inputs, in which the electrode impedances are equal and near zero. In real conditions, CMRR can drop significantly.

It is generally the case that a small (d.c.) offset voltage will exist between a pair of electrodes, whatever the materials used, and small offset voltages will be present within the input stages of an amplifier. These signals, once amplified, represent a significant proportion of the amplifier's dynamic range. (The dynamic range of an amplifier is the maximum input voltage that can be applied to the inputs at a given gain setting without overdriving the amplifier. If the amplifier is overdriven, the signal will be distorted or clipped.) A millivolt of d.c. offset becomes 20 volts after amplification of 20,000. Few electrophysiological amplifiers can produce an output as large as this, so no signal can be transmitted. To overcome this problem, most amplifiers are designed with capacitatively coupled stages to block the d.c. component of the signal. This is a.c. coupling. But suppose that the signal varies only very slowly with time; then it too will be blocked. Thus, a.c.-coupled amplifiers filter out the lower frequencies in a signal. A good example of this is the elimination of the c-wave of the ERG in most clinical recording situations. The inclusion of multiple a.c. coupling stages in an amplifier can lead to further quite specific problems.

If the input signal multiplied by the amplifier gain exceeds the dynamic range of any stage of the amplifier, then the output of that stage will be driven to the maximum voltage that can be delivered from the power supply. This will lock up the amplifier for a time that is determined by the time constant of that stage. The recovery will not begin until the source of the overdriving is removed. If a subsequent stage of the amplifier recovers more rapidly, then the output of the amplifier will return to the normal mean level, but no signal will be able to pass. Thus, each time an artifact (such as that produced by a blink during an ERG recording) occurs, it is possible, with poor amplifier design, to add a number of blank or zero records to the stored record, thus reducing the recorded signal amplitude. Pattern electroretinogram (PERG) recordings are particularly prone to this problem because of the small signal amplitude and large amplitude that eye movement artifacts present.

Filters

To reduce the amount of data that must be collected and processed and the difficulties of recording, the incoming signal is usually filtered to limit the frequency range to that of specific interest. For example, to record a PERG the band-pass of the amplifiers is set to 1.0–100Hz. The exact effect of these filters on the signal will be affected by a number of parameters associated with the filters that are used. The normally stated value of a filter is the cutoff point or shoulder frequency of the filter. It should be remembered that the filter frequency value that is normally quoted is the frequency at which the signal is reduced by 3dB, that is, reduced to 70% of its normal unfiltered amplitude. Therefore, a low-pass filter should be chosen that has a cutoff point sufficiently removed from the frequencies of interest. (A high-pass filter will remove frequencies below those of interest.) Another factor is the rate of attenuation with increase in frequency, or roll-off, of the filter. For an ideal filter, all frequencies above a given frequency would not be amplified, but most filters cannot achieve such a performance. For a simple resistor/capacitor (RC) network such as the decoupling capacitor used to block d.c. signals, there is an attenuation of 6dB per octave (roll-off). Cascading multiple stages can increase this comparatively shallow roll-off, although the complexity of the required circuitry increases dramatically with increase in the number of filter stages because buffers need to be inserted between stages. However, there is an added difficulty. At or near the frequency at which the filter begins to operate, the phase of the signal alters, sometimes dramatically, depending on the filter used. To produce a filter with an improved “roll off” characteristic without the complexity of cascaded RC stages and buffers, a tuned amplifier circuit or active filter is used. There are three major categories of active filter: Bessel, Butterworth, and Chebyshev. Of these, the Chebyshev has the better amplitude versus frequency characteristic, that is, a steep roll-off and sharp shoulder (see the graph in figure 16.3). The Bessel has the better time domain characteristics, that is, minimal phase change with frequency, although this is achieved at the cost of frequency domain performance. The Butterworth is a compromise between the two; it is generally used in audio equipment. For the purposes of electrophysiology, the time domain characteristics are important because the transient responses contain numerous frequency components, which will be affected to varying degrees by these phase changes. For example, the PERG P50 peak time is of clinical significance, and this can be changed by the injudicious choice of filter (see figure 16.3). The sacrifice of frequency domain performance becomes obviously worthwhile. Thus, the filter of choice is the Bessel filter.

Figure 16.3.  

The effect of low-pass filter settings on the PERG.

Another filter that is often found on modern equipment is the notch or line frequency filter. This is the inverse of a band-pass filter; that is, it attenuates only a very narrow band of frequencies, centered on-line or power supply frequency. Such a filter has a very high Q value, that is, a very steep roll-off. One problem with the use of this type of filter is that the line or mains frequency is relatively close to the frequencies that are to be recorded. This means that the phase changes that are introduced around the shoulders of this filter may well extend into the frequency band of the signal under investigation. Another problem that is encountered with notch filters but is often overlooked is the loss of an early warning of imminent electrode failure; that is, one does not pick up mains interference from an electrode in which the resistance is increasing. A preferable solution is simply to apply the electrodes correctly and set up the system properly, in which case a notch filter to remove mains frequency signals is not needed.

Real-time digital filters are becoming available, implemented using high-speed digital signal processors, a technology that was until recently very costly. While these filters are often used to mimic conventional filter designs, almost any filter configuration can be programmed. However, conventional analog filtering is still needed on the input stages.

Filters, correctly used, improve the signal-to-noise ratio. Another form of filtering, which behaves quite differently from these “real” filters, is conditioning of responses by software after they have been recorded. Various techniques are available, from three-point smoothing to Fourier decomposition of the signal, removal of power in selected frequencies, and signal reconstitution. The effect on phase information of this type of smoothing is not insignificant. It is not a substitute for predigitization filtering or good recording technique.

Analog-to-Digital Converter (ADC)

Once amplified to manageable levels and filtered of unwanted frequency components, the signal is converted from the analog into the digital domain for processing by the computer. The two main characteristics of the analog-to-digital converter (ADC) are the voltage resolution, defined in terms of the number of discrete levels (bits) available, and the frequency at which this subsystem can make these conversions.

The minimum acceptable resolution on modern equipment is 12 bits, giving 4096 levels. It should always be remembered that this is the maximum resolution of the converter. While this degree of resolution might appear excessive, it should be considered how much of this range is actually used for the response. Thus, suppose the effective amplifier input to give full range input to the ADC is set at ±0.5mV and a 2-µV PERG is to be recorded. This signal voltage is only twice the voltage separating each level of a 10-bit ADC; therefore, only 2 bits of resolution are available for the waveform to be recorded.

The temporal resolution or conversion rate of an ADC subsystem is quoted in terms of its maximum throughput. With a good-quality converter, this will represent the true speed of operation, but this is not necessarily achievable under program control. The maximum signal frequency that the unit can convert is sometimes quoted as the data throughput rate divided by 2. This is the case only for a signal that is both digital in nature and synchronous with the conversion clock. Analog signals require more than the maximum and minimum points to resolve the waveform in a satisfactory fashion.

A common source of error with converters is aliasing. This occurs when the sampling frequency is inappropriate for the signal under examination. This can result in a signal appearing to be of a totally different frequency, that is, aliasing (figure 16.4).

Figure 16.4.  

Aliasing. The light line represents the actual signal, a high-frequency sine wave; the vertical bars represent the sampling points; and the heavy line is the perceived signal.

Because most systems use a single ADC, another device is required in multiple-channel systems; this is the sample and hold amplifier (figure 16.5). The purpose of this component is to freeze in time the input to the ADC across all channels so that, as the converter scans across the inputs, the channels are in effect, all sampled at the same time. Without the sample and hold, the output would be slewed by the channel number multiplied by the channel read time.

Figure 16.5.  

The sample and hold amplifier.

Signal Extraction

At this point, the digital signal may still be concealed within a considerable amount of noise. To extract the signal from the noise, some form of signal extraction technique is required. A number of such techniques are available. Phase-sensitive (lock-in) amplifiers extract the level of the fundamental frequency from the incoming waveform. If a stimulus is repetitive and of a frequency such that the visual system can respond to each stimulus, the incoming voltage has repetitive changes. Fourier theory states that this signal can be decomposed into a number of different sine waves (see chapter 31). In a lock-in amplifier, the input is multiplied by a sine wave of exactly the input frequency. In most equipment, there are two such amplifiers, and the internally generated sine wave in one is 90 degrees out of phase with the other. The outputs of the two amplifiers are squared and added, and this output is a measure of the amplitude of the input signal (because of the identity cos²θ + sin²θ = 1), either of its fundamental or of the second harmonic (for parts of the visual system that give ON-OFF responses). This is a powerful method and can give a rapid result, because all information about the actual waveform of the response is lost. It is an example of the general technique of cross-correlation. Full-scale Fourier analysis can now be used to similar effect by recording continuously into memory and analyzing off-line after the period of stimulation ends. The computer returns the result almost instantaneously. The power and phase that are contained at a particular frequency can be analyzed. (The fundamental here is the total duration of the record.) Such techniques have pitfalls for the inexpert user. For example, the slightest change in stimulus or internally generated sine wave frequency can have disastrous results, and when stimuli are generated by the computer, glitches occur owing to the vagaries of operating systems. Again, if the stimulus is produced on a slow raster display and the response comes from the fovea, eye movements can alter the real frequency of the stimulus. For further analyses, see chapters 18 and 34.

The most commonly used technique is that of signal averaging. This relies on the fact that the response to be recorded is synchronous with the stimulus that is used to evoke it, whereas the background noise is random with respect to the stimulus. The technique is to record for a fixed time interval following and triggered by the stimulus, for example, a pattern reversal. When the next reversal occurs, another time interval is recorded and added to the first, and the result is divided by the number of time intervals (sweeps). The noise being a random with respect to time, the stimulus should, as the process is repeated, cancel out, leaving only the signal evoked by the stimulus. The extent to which averaging enhances the signal-to-noise ratio is given as 1/ N , where N is the number of sweeps or repetitions. From this formula, it can be seen that the law of diminishing returns applies.

Artifact Rejection

One problem that can occur with signal averaging is the contamination of the “good” recording with artifacts caused, for example, by a blink or other movement. The solution to this problem is to utilize an artifact rejection system, whereby before data are passed into the signal averager, they are examined to ensure compliance with preset conditions, typically, given that the signal amplitude does not exceed preset limits at any point. If the conditions are met, then the data are passed to the averager; if not, then the entire sweep is rejected. A wide range of rejection criteria have been tried; the most successful in general use and found on most equipment is the simplest, and is given above. The upper and lower preset limits should be variable by software. Other limits sometimes encountered include a maximum rate of change of voltage.

Control Software

Two main approaches to this vital component of the recording system exist, to allow either full control of every parameter for both stimulation and data acquisition or a series of preprogrammed tests with minimal opportunity for the operator to vary the parameters. Which option is preferred will depend on the application. An ideal solution is to have both options available so as to encompass the requirements of routine clinical examination and research.

The recording from the patient should be stored either on the local machine's fixed disk or on a network device in larger installations. Data is normally held in database form, which enables a record of all settings used during the recording to be stored attached to the traces, along with any annotation that is made during the recording. All this information should be available on printed records produced by the system, along with detailed cursor displays and measurements of timing and amplitude. An advantage of storing data in database format rather than discrete patient files is the ability to search a large number of records on specific criteria. Raw data files that can be exported for complex manipulation with third party software is a distinct advantage, although some manufacturers work with impenetrable proprietary data formats that prevent this. Adequate data backup provision on removable media is essential.

Commercially Available Electrophysiology Systems

Table 16.1 lists many of the currently known commercial manufacturers of visual electrophysiology systems. A number of commercial systems were originally designed for electromyographic (EMG) use and have had evoked potential capabilities added and then visual evoked potential capabilities added. A small number are designed specifically for visual electrophysiology. This situation is brought about by the very high cost of designing, manufacturing, and then marketing systems that comply with regulatory demands and the relatively small number of units carrying out visual electrophysiological examinations worldwide.

In selecting equipment, many factors need to be taken into account. High on the list of selection criteria should be the availability of technical support from the manufacturer to deal with basic system questions, setup, resolution of software bugs, and assistance with developing new protocols that are specific to the needs of the laboratory. The frequency and availability of software and equipment upgrades are important in light of the continuous evolution of testing protocols for specific disease entities. Should a system that is being considered not immediately fulfill your laboratories current and foreseeable requirements, it is advisable to treat any promise of future development with a degree of skepticism.

Systems must be user-friendly, reliable, and cost-effective, although cost is surprisingly less of a concern if it is eclipsed by the power of the system. Adherence to International Society for Clinical Electrophysiology of Vision (ISCEV) standards for all packaged testing protocols is obviously a requirement for any clinical laboratory, although the ability to exceed these standards by a significant margin is advisable.

Appropriate calibration of stimuli so that they conform to laboratory and international standards is a major concern, and some manufacturers offer calibration services and equipment for this purpose. While manufacturers are encouraged to incorporate calibration devices that automatically execute on system startup and alert users to potential calibration failures, few do, and additional independent calibration equipment is preferable and should be included in budget proposals.

Finally, data should be stored in a form that is easily exportable to other software packages. Some systems are designed with proprietary data formats, which make the raw data almost impossible to extract for further analysis.

As with any major purchase, the opinions of one's more experienced colleagues will be helpful and generally freely available.

Building Your Own

It is a relatively easy task to assemble a data acquisition system from individual components, provided that one possesses a moderate level of electronics and computer programming skills. The major advantage of such home-built systems is that they are relatively cheap to assemble and they can be designed to meet the specific needs of a laboratory or clinical facility. However, a major disadvantage is that the writing and debugging of the software code that makes the systems function is extremely time-consuming and laborious, frequently evolving over many years. Some device manufacturers will provide sophisticated software function calls and/or prepackaged nonspecific software code that can make the job of programming much easier. In addition, any home-built system that will be used in a clinical setting must meet all safety and locally mandated regulatory regimes.

A listing of some of the major manufacturers of specific hardware and software components needed for a typical electrodiagnostic system was given earlier in the chapter in table 16.1.