Quality of the average response

Traditionally, evoked responses have been obtained by averaging a fixed number of sweeps. This means that in some cases test time has been wasted in obtaining sweeps which were not necessary whilst in other cases diagnostic decisions have been based upon poor waveforms for which more sweeps were needed. More flexible, adaptive procedures could be implemented if there was a rule for deciding how many sweeps are required to obtain a certain ‘quality of response’. Quality estimation is based upon the signal/noise ratio and various methods of estimating this have been tried. It is important as this will permit a greater reliance to be placed on the diagnostic interpretation of good quality responses and, if the estimation is computed at the same time as the averaging process occurs, it would allow the operator to obtain just enough sweeps to give a certain quality of waveform. The resultant improvement in interpretation and saving in test time makes a significant advance to clinical practice.

The basis for most quality estimates is the measurement of the noise variance. Schimmel (1967) proposed the ‘± average’ method to estimate this. For each pair of sweeps the second is subtracted from the first to cancel out the response component and leave only the noise. This is then stored in a separate memory area to that which contains the normal average. Wong & Bickford (1980) used this method as the basis of a quality estimator and Elberling & Don (1984) further developed the technique, making the estimator more robust against large amplitude, low frequency noise components. Several commercially available evoked response systems are beginning to implement such methods and this should improve the quality of clinical recordings.

Weighted averaging

Clinical experience indicates that the usual averaging procedure does not always perform well. Most clinicians with experience in this field will have come across cases in which increasing the number of sweeps can even lead to a less well-defined response. This occurs when the background noise is not ‘stable’. Stationarity or ‘stability’ of the noise is fundamental to the application of averaging but as the patient changes his state of arousal so the noise can change. Hoke et al (1984) modelled this non-stationary behaviour and derived a weighted averaging technique which gave an improvement over normal averaging for such a process. Elberling & Wahlgreen (1985) developed this idea and, using Bayesian statistics, described an elegant and practical method of implementing weighted averaging. The averaging process is carried out in blocks of a relatively small number (approximately 250) sweeps. For each block the mean wave form and a noise variance are calculated and each block is weighted according to its noise variance. Thus blocks that come from noisier portions of the record contribute less to the final average than do the blocks that come from quiet sections of the record. In this way, short periods of high activity, or transients that contaminate the response, play very little part in the final average and this technique can give the same quality of response in a shorter time than normal averaging.

Further techniques

Research is continuing in many other areas and these are still remote from clinical practice and so will be mentioned only briefly.

The SQUID (Superconducting Quantum-Interference Device) is the device used to record the magnetic field produced by neural and myogenic activity. A conceivable advantage of magnetic as opposed to electrical recording lies in the possibility of directly estimating the generator site; the magnetic, tonotopic representation of the auditory slow vertex response has been reported by Pantev et al 1986).

Fundamental insight has been obtained from deconvolution techniques in which the normal click-evoked brainstem response is deconvolved with a unit response to give the compound impulse response of the brainstem generators. This mathematical exercise is useful because it is a means of effectively removing the changes to the response caused by peripheral hearing loss. Despite the advantages of this technique, its complexity has meant that it has not been applied as a routine clinical test.

Using three orthogonal electrode pairs, Williston et al (1981) recorded ABR data to produce a three dimensional representation in voltage/time space. Certain segments of this plot lie in a plane and the orientation of the plane can be altered by disorders of the auditory system. Further research is expected to refine the interpretation of the data and to investigate the effects of pathology.

Using a montage of 16 or more electrodes placed over the scalp, the evoked response amplitude at any instant can be mapped as isopotential lines. The potential at each electrode is used to interpolate the potentials at the points between the electrodes and, using colours to represent the potential value, a colour map of the evoked response distribution over the surface of the scalp can be generated. Such mapping techniques have been used both to gain fundamental knowledge and in clinical application (Morihisa et al 1983). More recently a technique of ‘isochronic’ mapping has been applied to the auditory brainstem responses (Thornton et al 1990). In this technique, lines of equal time, rather than isopotential lines, are plotted. The map therefore shows the latency of a particular peak in the response map over the scalp.