estimating audiometric thresholds using auditory steady ...were presented simultaneously, four to...

J Am Acad Audiol 16:140–156 (2005)

140

*Rotman Research Institute, Baycrest Centre for Geriatric Care, University of Toronto; †Cuban Neurosciences Research Center,Havana, Cuba

Terence W. Picton, Rotman Research Institute, Baycrest Centre for Geriatric Care, 3560 Bathurst Street, Toronto, M6A 2E1,Canada; Phone: 416-785-2500, ext. 3505; Fax: 416-785-2862; E-mail: [email protected]

Estimating Audiometric Thresholds UsingAuditory Steady-State Responses

Terence W. Picton*Andrew Dimitrijevic*Maria-Cecilia Perez-Abalo†Patricia Van Roon*

Abstract

Human auditory steady-state responses (ASSRs) were recorded using stimulusrates of 78–95 Hz in normal young subjects, in elderly subjects with relativelynormal hearing, and in elderly subjects with sensorineural hearing impairment.Amplitude-intensity functions calculated relative to actual sensory thresholds(sensation level or SL) showed that amplitudes increased as stimulus intensityincreased. In the hearing-impaired subjects this increase was more rapid atintensities just above threshold (“electrophysiological recruitment”) than athigher intensities where the increase was similar to that seen in normalsubjects.The thresholds in dB SL for recognizing an ASSR and the intersubjectvariability of these thresholds decreased with increasing recording time andwere lower in the hearing impaired compared to the normal subjects. After 9.8minutes of recording, the average ASSR thresholds (and standard deviations)were 12.6 ± 8.7 in the normal subjects, 12.4 ± 11.9 dB in the normal elderly,and 3.6 ± 13.5 dB SL in the hearing-impaired subjects.

Key Words: Auditory steady-state responses, objective audiometry, physiologicalthresholds, signal-to-noise ratio

Abbreviations: AD = analog-digital; ASSR = auditory steady state response;DA = digital-analog; EEG = electroencephalogram; MASTER = multiple auditorysteady-state response; PTA = pure tone average (for 500, 1000, 2000, and 4000Hz); SL = Sensation Level; SNR = signal-to-noise ratio

Sumario

Se registraron respuestas auditivas de estado estable (ASSR) en humanos,usando una tasa de estímulos de 78-95 Hz en sujetos jóvenes con audicónnormal, en sujetos ancianos con una audición relativamente normal, y en sujetosancianos con un trastorno auditivo sensorineural. Las funciones de amplitud-intensidad calculadas en relación con los umbrales auditivos reales (nivel desensación o SL) mostraron que las amplitudes aumentan conforme seincrementa la intensidad. En los sujetos hipoacúsicos este incremento fue másrápido en intensidades justo por encima del umbral (“reclutamientoelectrofisiológico”) que a intensidades mayores, donde el incremento erasimilar al visto en los sujetos normales. Los umbrales en dB SL para reconoceruna ASSR, y la variabilidad inter-sujetos de estos umbrales, disminuyóconforme aumentó el tiempo de registro y fue menor en los sujetos hipoacúsicos,comparada con los normales. Luego de 9.8 minutos de registro, los umbralespromedio de las ASSR (y sus desviaciones estándar) fue de 12.6 ± 8.7 en lossujetos normales, 12.4 ± 11.9 dB en los ancianos normales y 3.6 ± 13.5 dBSL en los sujetos hipoacúsicos.

Recording human auditory steady-stateresponses at different intensities canprovide an objective assessment of

audiometric thresholds (reviewed in Picton etal, 2003). Present practice records theresponses to tones that are amplitudemodulated, or amplitude and frequencymodulated, at rates of 70–110 Hz. At theserates the responses are not attenuated bysleep (Cohen et al, 1991; Picton et al, 2003)and can be recorded in young infants(Rickards et al, 1994; Lins et al, 1996; Savioet al, 2001; Cone-Wesson et al, 2002a, 2002b,2002c; Luts et al, 2004).

The accuracy with which behavioralthresholds can be estimated from thephysiological responses depends on how theresponses are recorded. The simplesttechnique is to estimate behavioral thresholdsas equivalent to the threshold for recognizingthe physiological response as significantlydifferent from noise. In general suchestimated thresholds will be higher than theactual behavioral thresholds by amounts thatvary between 5 and 40 dB (Rickards et al,1995; Lins et al, 1996; Herdman and Stapells,2001, 2003; Perez-Abalo et al, 2001;Dimitrijevic et al, 2002; Rance and Briggs,2002). The difference between estimated andactual thresholds varies with severalparameters, among the most important ofwhich are the frequency of the sound, thedegree of hearing loss, the age of the subjects,and the duration of the recording.

Our contention is that these parametersbasically depend on the signal-to-noise ratio(SNR) of the recording. The amplitude of aresponse varies with the carrier frequency,responses at frequencies 1000 and 2000 Hzbeing larger than responses to lower andhigher audiometric frequencies (see John etal, 2001b). Physiological thresholds are closerto the behavioral thresholds for patients witha sensorineural hearing loss than for normalsubjects (e.g., Dimitrijevic et al, 2002; Ranceand Briggs, 2002). When a sensorineuralhearing loss is present, the amplitude of the

response likely increases more rapidly thannormal in the first 10–20 dB above threshold,making the response easier to recognize nearthreshold. This presumed amplitude-intensityfunction has not been clearly demonstrated,although Dimitrijevic et al (2002) reportedthat for stimuli of equal hearing level (HL)the response in patients with sensorineuralhearing loss is larger than in normal subjects.Age can affect the amplitude of both theresponse and the noise. In infants and inelderly subjects the noise levels are oftenhigher than in young adults, and in infantsthe responses are smaller (Lins et al, 1996;Savio et al, 2001; John et al, 2004). When theduration of the recording is longer, theresidual noise of the electroencephalogram(EEG) is less (John and Picton, 2000; Luts etal, 2004), making small near-thresholdresponses easier to recognize.

In order to evaluate this idea, we recordedASSRs (auditory steady-state responses) innormal subjects and in hearing-impairedsubjects at different intensities abovebehavioral threshold and for differentdurations. Our hypotheses were that thephysiological thresholds would be closer tobehavioral thresholds and less variable for thefrequencies 1000 and 2000 Hz compared tohigher or lower frequencies, for the hearing-impaired subjects compared to the normalsubjects, and for longer compared to shorterdurations of recording. In the hearing-impaired subjects, we recorded at 5 dB stepsin order to delineate any abnormality of theamplitude-intensity function.

METHODS

Subjects

We examined three separate groups ofsubjects. Ten normal young subjects (meanage 25 years, range 19–31, 5 male) wererecruited from laboratory personnel. All hadthresholds equal to or less than 20 dB HL at

Threshold Estimation Using ASSRs/Picton et al

141

Palabras Clave: Respuestas auditivas de estado estable, audiometría objetiva,umbrales fisiológicos, tasa señal/ruido

Abreviaturas: AD = analógico-digital; ASSR = respuestas auditivas de estadoestable; DA = digital-analógico; EEG = electroencefalograma; MASTER =respuestas auditivas múltiples de estado estable; PTA = promedio tonal puro(para 500, 1000, 2000 y 4000 Hz); SL = nivel de sensación; SNR = tasaseñal/ruido

the frequencies 500, 1000, 2000, and 4000 Hz.Their pure-tone averages (PTA) across thefrequencies 500, 1000, 2000, and 4000 Hzaveraged 3 with a range of -1 to 9 dB. Tenhearing-impaired subjects (mean age 78 years,range 64–86, 4 male) replied to a noticerequesting subjects posted in the audiologyclinic at Sunnybrook and Woman’s CollegeHealth Sciences Centre, or volunteeredthrough our subject pool. These subjects wereselected to have a moderate or severesensorineural hearing loss. Their average PTAwas 51 with a range of 34–74 dB. Since thesesubjects were older than our normal subjects,we also examined a group of 10 elderly subjects(mean age 69 years, range 61–77, 2 male)from our volunteer pool. These subjects hadnormal hearing or a mild high-frequencysensorineural loss (untreated with hearingaids). Their average PTA was 15 with a rangeof 2 to 36 dB. Table 1 shows the averagebehavioral thresholds at each audiometricfrequency for each group of subjects.

Stimuli

Stimuli were created and presented usingthe multiple auditory steady-state response(MASTER) system (Lins and Picton, 1995;John et al, 1998; John and Picton, 2000; seealso www.mastersystem.ca). Eight stimuliwere presented simultaneously, four to eachear. Carrier frequencies were 500, 1000, 2000,and 4000 Hz in each ear. The stimuli weremodulated using exponential modulationwith an exponent of 2 (John et al, 2002a). Forthe main experiments, the stimuli weremodulated at frequencies between 78 and 95Hz, with each carrier frequency associated

with a unique modulation frequency. Thenormal young subjects were also tested usingmodulation frequencies between 110 and 125Hz. (The rationale of this part of theexperiment was to determine whether the78–95 Hz rates were indeed optimal orwhether responses might be more readilyrecognized at faster rates where the noiselevels were lower). Table 2 shows the exactmodulation frequencies. The modulationfrequencies were intercalated so that thesame carrier frequencies in each ear wereadjacent and increased with increasingcarrier frequency beginning with the rightear. The stimuli were digital-to-analog (DA)converted at a rate of 32 kHz and routed toa Grason Stadler Model 16 audiometer wherethey were amplified to a calibration intensity,attenuated to achieve the desired intensitylevels, and presented using Etymotic-3Ainsert earphones. Although the MASTERsystem simultaneously presented four stimulito each ear, each stimulus was separatelycalibrated (accuracy to within 2 dB) using aBruel and Kjaer DB 0138 coupler toequivalent intensity measured in root-mean-square sound pressure level (SPL). Thecombined stimulus was 7 dB more intensethan the individual stimuli.

In the normal subjects (both young andelderly) the stimuli were presented at levelsof 60 dB SPL and decreased in intensity by10 dB steps until no responses wererecognized as significant or until 10 dB SPL.Normal subjects were examined on twoseparate days, one for the 78–95 Hzmodulation rates and one for the 110–125 Hzrates. In the hearing-impaired subjects, theintensity began at 80 dB SPL and decreased

Journal of the American Academy of Audiology/Volume 16, Number 3, 2005

142

Table 1. Behavioral Thresholds (mean and range in dB HL)

Group Frequency (Hz)

500 1000 2000 4000

Young 6(-5 to 20) 4(-5 to 15) 4(0 to 10) -2(-10 to 5)

Elderly 15(5 to 20) 11(0 to 30) 12(0 to 30) 22(0 to 55)

Hearing Impaired 39(15 to 65) 48(5 to 65) 57(35 to 80) 60(35 to 95)

Table 2. Stimulus Modulation Frequencies (Hz)

Ear Left Right

Carrier 500 1000 2000 4000 500 1000 2000 4000

78–95 Hz 80.1 85.0 89.8 94.7 78.1 83.0 86.9 91.8

108–125 Hz 110.4 115.2 120.1 125.0 108.4 113.3 117.2 122.1


143

in 5 dB steps. This required two separaterecording sessions.

Recordings

Auditory steady-state responses wererecorded from an electrode placed at thevertex, using a neck electrode as a referenceand an electrode placed on the left clavicle asground. Subjects slept or drowsed in areclining chair, located within a darkenedsound-attenuated chamber during therecordings. The EEG was collected using afilter bandpass of 1 to 300 Hz, a gain of50,000, and an analog-digital (AD) rate of1000 Hz. The EEG was continuouslymonitored to ensure that the subject wasrelaxed, that is, showing little fast muscleactivity. Individual EEG epochs of 1024 pointseach (1.024 seconds) were rejected if theycontained any value that exceeded ±80 µV.Sixteen individual data epochs of 1024 pointseach were collected and linked together intosweeps lasting 16.384 seconds. As each sweepwas completed it was combined with arunning average sweep using weightedaveraging based on the activity in the signalbetween 70 and 110 Hz (John et al, 2001a).The averaged sweep was analyzed using aFast Fourier transform (FFT). The amplitudeof the steady-state response to a given carrierfrequency was equal to the amplitude at thefrequency of modulation in the resultingspectrum. This amplitude was compared tothe amplitudes in adjacent regions of thespectrum (60 frequencies above and 60frequencies below the response frequency)using an F-ratio with 2 and 240 degrees offreedom (Lins et al, 1995). A response wasjudged as significantly different from residualEEG noise at a criterion of p < 0.05. For eachexperimental condition of our experiments,36 sweeps, or approximately 10 minutes ofdata was collected. All data were stored todisk and then analyzed offline after 1, 3, 6,12, 24, and 36 sweeps.

Analyses

Two decisions were necessary beforeevaluating the amplitude-intensity functions.The first decision was whether to base theintensity on the normal thresholds for eachcarrier frequency (hearing level or HL) or oneach subject’s own threshold (sensation levelor SL). Since our intent was to show the

response increased more rapidly fromthreshold in subjects with a hearingimpairment, we decided to use SL. In orderto convert our intensities from SPL to SL weused the SPL of the stimulus minus the SPLof the tone used for the audiometricevaluation. In order to allow us to combinedata across subjects these differences wererounded to the nearest 5 dB. However, inthe normal subjects, where the intensitieswere changed in 10 dB steps, this would givemissing data at many intensities. Thisapproach gave informative graphs about theeffect of hearing loss on the responses butmade it impossible to conduct statisticalanalyses. We therefore used SPL to measurethe effects of ear, intensity, frequency andage. This analysis used a four way ANOVAwith repeated measures within the factorsear, intensity and frequency. A second analysiscompared the young and elderly with theelderly shifted 10 dB to compensate for theaverage change in the PTA. In this analysisthe response of the elderly subjects at 60 dBSPL was compared to the response of theyoung subjects at 50 dB SPL, and this wasrepeated down to 20 dB in the elderlycompared to 10 dB in the young. A thirdanalysis compared the responses in the youngsubjects across the two different modulationfrequencies (78–95 versus 108–125 Hz).Geisser Greenhouse criteria were used whenappropriate to assess the repeated measureseffects.

The second decision was how to assessamplitudes when the response was notsignificant. One way is just to use whateveramplitude was measured. In this approachthe amplitude decreases with decreasingintensity and then flattens off at the noiselevel of the recording for responses that arebelow the physiological threshold. The secondapproach is to consider the amplitude of anonsignificant response as zero. This maybe inappropriate when the noise level isparticularly high for one recording but isgenerally still preferable to taking the noiselevel as the amplitude of the response. Weopted for the second approach.

The noise level of the recordings wasdetermined as the mean amplitude of theresponse in the ±60 bins of the spectrumadjacent to the response bin (±3.7 Hz). AnANOVA compared groups with repeatedmeasures for carrier-frequency, ear, andintensity. Since there were incomplete data

at the lower intensities for the hearing-impaired subjects, this only involved levelsof 50, 60 and 70 dB SPL.

Physiological thresholds were determinedusing two rules. The first was that thethreshold was the lowest intensity at whicha significant response was detected, providedthat the response was insignificant at nomore than one higher intensity. The secondwas that if no threshold was determinedaccording to the first rule then the thresholdwas set at 10 dB above the highest intensityused to elicit the response. Since the stepsize for intensity in the hearing-impairedpatients was smaller than in the othersubjects, we calculated the thresholds inthese subjects using 10 dB step sizes (either5-15-25 … or 10-20-30 …) when makingcomparisons across groups. The residualEEG-noise levels in the recordings were lowerfor the young subjects compared to the othertwo subject groups, and this could alter anycomparison of threshold estimation. Wetherefore decided to compare physiologicalthresholds for the hearing-impaired subjectswith those of the normal elderly. Thedifference between physiological andbehavioral thresholds were thus evaluatedusing an ANOVA with two subject groups(normal elderly, hearing impaired) andrepeated measures for different times ofrecording (or number of sweeps averaged),ear, and carrier frequencies. A further ANOVAcompared the thresholds for the differentmodulation frequencies in the young subjects.

There is no readily available method totest for changes in intersubject variability ina multifactorial ANOVA. In order to get somesense of changes in this variability, weselected data for the responses at 2000 Hz,combined data across the two ears, andestimated Levene’s statistic to evaluatewhether there was a significant change invariance between the data obtained afteraveraging 6 sweeps (1.6 minutes) and after36 sweeps (9.8 minutes). This analysis wasperformed for selected intensities for theamplitude and noise measurements and forthe estimated physiological thresholds

RESULTS

Amplitude-Intensity Functions

Typical responses are shown in Figures

1 and 2. Figure 1 shows ASSRs recorded atintensities ranging from 20 to 70 dB SPLfrom a single young subject at two differentmodulation bands, 78–95 Hz, and 108–125 Hzafter the full recording time (36 sweeps, 10minutes) at each intensity. Figure 2 showsASSRs recorded at intensities ranging from20 to 70 dB SPL from a single elderly subjectwith normal hearing sensitivity after 6sweeps (left side) and after 36 sweeps (rightside). The mean amplitudes for the responsesacross the groups are shown in Figure 3. Theinitial ANOVA (comparing responses in theyoung and elderly groups at the same SPL—the data on the left side of Figure 3) showeda main effect of group (F = 9.0; df = 1,18; p <0.01) and intensity (F = 111.3; df = 5,90; p <0.001) and a group by intensity interaction(F = 4.3; df = 5,90; p < 0.01). The amplitudewas larger in the young than the elderlysubjects, and for higher intensity comparedto lower intensity, with the interaction beingcaused by the group effect being larger athigher intensities. The 500 Hz response wassmaller than the others, and this effect wasmore prominent at low intensity (frequencyand intensity interaction F = 7.4; df = 15,270;p < 0.001). Since we found no effects of earor interactions with ear, the responses werecollapsed across ears in Figure 3. When theresponses in the young were compared withthe elderly response to stimuli 10 dB lower,the group effect and group interaction withintensity was no longer significant, but thefrequency effect and frequency by intensityinteraction remained. In the young subjects,the amplitudes of the responses were muchsmaller at 108–125 Hz compared to 78–95 Hz(graphs in the upper half of Figure 3) (F = 26.0; df = 1,9; p < 0.001), especially athigher intensity (intensity by rate interactionF = 11.3, df = 5,45; p < 0.01).

Results in Hearing-Impaired Subjects

Responses for two different hearing-impaired subjects are shown in Figure 4. Theseresponses were those obtained after averaging36 sweeps. The average amplitudes in thehearing-impaired subjects are shown in thebottom right of Figure 3. The main findingswere that the responses were smaller at lowerintensities, which were subthreshold, and thatthis difference was greater for the highercarrier frequencies, where the thresholds werehigher. At the higher intensities where the


144


145

Figure 1. ASSRs in a normal young subject. On the left are plotted the responses to the ASSRs using stimulimodulated in the frequency range of 78–95 Hz, and on the right are the responses for 108–125 Hz. The responseswere averaged over 36 sweeps. Dotted lines separate the results at the different carrier frequencies. Theresponses identified as significant are indicated by the triangles. The filled triangles indicate the responses forright ear stimuli, and the open triangles are for left ear stimuli. The carrier frequencies increase with increas-ing modulation frequency (Table 2), and thus the identified responses beginning on the left are to 500 Hz in theright ear, 500 Hz in the left ear, 1000 Hz in the right ear, etc. The responses at 78 and 80 Hz (for 500 Hz carri-ers in the right and left ears, respectively) are not recognized at 40 dB SPL but then become recognizable at 30dB. According to our rules, the thresholds would have been 30 dB SPL. The responses at the faster modulationfrequencies are smaller. However, since the noise level is also smaller, the responses are recognized down to sim-ilar thresholds.

Figure 2. Effects of averaging. This shows the response in a normal elderly subject, as shown after 6 sweepsand 36 sweeps were averaged. The residual EEG-noise levels decrease with increased averaging. This makesit easier to recognize responses at low intensities, thus leading to lower physiological thresholds. Physiologi-cal threshold levels (determined according to the rules described in the text) are shown with asterisks. Theresponses indicated with “?” are likely false positives. The response at 1000 Hz in the left ear at 20 dB SPL islikely a false positive but according to our rules is judged suprathreshold. Even granting this value, the phys-iological thresholds are on average 15 dB lower after averaging more sweeps.

stimuli were above threshold, however, theamplitudes of the responses were similar tothose of the normal elderly (e.g., compare dataat 80 dB SPL in the hearing impaired to thedata at 70 dB SPL in the elderly).

Sensation Levels

In order to compare the data across thedifferent subjects, we readjusted the intensitiesto SL by subtracting the behavioral thresholdsfrom the actual physical intensities. Sampledata from individual subjects (one normalhearing and one hearing impaired) are plottedin Figure 5. The mean data (for each carrierfrequency and for each of the three subjectgroups) are plotted in Figure 6. The graph inthe upper right of Figure 6 collapses the dataacross carrier frequencies and plots all threegroups together. Since the behavioral thresholdswere obtained in 5 dB steps and since theASSRs were only obtained at 10 dB steps in thenormal young and elderly subjects, it was notpossible to evaluate these data using an ANOVAmodel. The mean data were plotted only if fouror more values contributed to the mean.

Residual EEG Noise

As can be seen in Figure 2, the level ofthe residual noise in the recording—the

amplitudes at bins in the spectrum that didnot contain responses—decreased with anincrease in the number of sweeps averaged.The ANOVA showed a significant main effectof group (F = 5.5; df = 2,27; p < 0.01) with theyoung subjects having lower levels than theother two groups. The noise levels weresignificantly lower for lower intensities, forhigher numbers of sweeps, for higher carrierfrequencies, and for the left ear as opposedto the right ear. Graphs showing the effectsof number of sweeps, intensity, and groupare shown in Figure 7. The effects offrequency and ear were too small to be easilyvisible on such graphs. The changes in thevariability of the noise from averaging 6sweeps to averaging 36 sweeps are shown inTable 3 together with the changes in theresponse amplitude. We present only thedata for 2000 Hz at selected intensities.

Thresholds

The physiological thresholds decreasedwith an increase in the number of sweepsaveraged. This is illustrated for actual ASSRsin Figure 2. The mean difference betweenthe physiological thresholds and thebehavioral thresholds (i.e., the physiologicalthresholds expressed in dB SL) are graphedin Figure 8, which shows the effects of carrier


146

Figure 3. Amplitude-intensity functions. Amplitudes are plotted against intensity in dB SPL for the differentgroups (young, elderly, and hearing impaired) for the 78–95 Hz responses. Also shown are the amplitudes forthe 108–125 Hz responses in the young subjects. The average amplitudes at each of the carrier frequencies arecollapsed across ears.


147

Figure 4. ASSRs in two hearing-impaired subjects. The behavioral thresholds (in dB SPL) are given at the topof the graph. The responses were averaged over 36 sweeps. For the subject on the left of the figure, the physi-ological threshold at 78 Hz (500 Hz to right ear) is 60 dB, which is 25 dB higher than the behavioral thresholdat 35 dB SPL. The physiological threshold at 95 Hz (4000 Hz to right ear) is 60 dB, which is the same as thebehavioral threshold. Note that the amplitude scales are different since the responses were much larger in thesubject whose ASSRs are plotted on the right. (The general noise levels of the residual EEG were similar forthe two subjects. The response at 30 dB to 2000 Hz in the left ear (identified with “?”) is a false positive response.

Figure 5. ASSR amplitudes relative to sensation level (SL). This figure shows the response amplitudes (78–95Hz) of two subjects (one ear each) in relation to intensity as plotted in dB SL. The young subject on the left hasnormal hearing (thresholds in dB HL are given at the top of the graph), and the amplitudes increase with increas-ing intensity at approximately the same rate regardless of intensity or frequency. The subject on the right hasa hearing impairment. The amplitudes increase rapidly immediately above threshold and then continue to increaseat a lesser rate with increasing intensity.

frequency in the three groups of subjects andthe overall effects of subject group in theupper right graph. The ANOVA (performedonly using the 10 dB stepped data for hearingimpaired) showed a main effect of group (F= 19.7; df = 1,18; p < 0.001) with the hearingimpaired showing lower thresholds than thenormal elderly subjects. The within-subjectanalyses showed the anticipated main effectof number of sweeps (F = 123.2; df = 5,90; p< 0.001) and an interaction between subjectgroup and number of sweeps (F = 6.9; df =5,90; p < 0.01). This interaction was causedby the threshold decrease with increasing

number of sweeps being less for the hearingimpaired than for the other subjects. A maineffect of frequency (F = 16.3; df = 3,54; p <0.001) was related mainly to the threshold at500 Hz being higher than at the otherfrequencies. A second ANOVA compared thephysiological thresholds in the young subjectsacross the different modulation rates (78–95versus 108–125 Hz) showed no significantdifference between the modulation rates.

Table 4 shows the results of our analysisof the intersubject variability in thresholdestimation. The standard deviations ofthreshold estimations are tabulated after 6


148

Figure 6. Amplitudes relative to SL. This set of graphs show the mean data plotted for all frequencies and sub-ject groups. ASSRs were obtained in 10 dB steps in the normal-hearing subjects, but since SL was calculatedin 5 dB steps, the data could be plotted in 5 dB steps. For simplicity, the graph only uses symbols at every 10dB step. The graph in the upper right of the figure shows the data averaged across groups.

Table 3. Variability of Response Amplitude and Residual EEG Noise (nV)

Measurement Condition 6 Sweeps 36 Sweeps Probability*Mean SD Mean SD F-test Levene

Response Young 60 dB SPL 75.6 21.1 72.9 20.6 ns ns

Elderly 70 dB SPL 60.5 31.3 58.2 18.1 ns 0.051

Hearing Impaired 80 dB SPL 57.3 27.6 53.8 23.8 ns ns

Noise Young 60 dB SPL 17.5 4.1 8.1 2.6 <0.001 0.002

Elderly 70 dB SPL 30.6 9.5 11.1 3.3 <0.001 0.002

Hearing Impaired 80 dB SPL 33.3 10.8 12.6 3.7 0.001 <0.001

* Probability of the null hypothesis. The F-test evaluates the difference between the means, and the Levene test evaluatesthe difference between the variances. Only values below 0.10 are tabulated.

and 36 sweeps. In addition to the decreasingmean threshold, the intersubject variabilityof the physiological threshold decreased withan increasing number of sweeps. This wasmore often significant for the normal-hearingsubjects (young and elderly) than for thehearing-impaired subjects.

DISCUSSION

Prefatory Comments

The results clearly demonstrated thatthe physiological thresholds were furtheraway from behavioral thresholds for 500 Hzcarriers than for the 1, 2, or 4 kHz carriers.We had hypothesized that the thresholds forthe 4 kHz carrier would also be elevated, butthis was not the case. Physiological thresholds


149

Figure 7. Levels or residual EEG noise during ASSR recordings. Both graphs show how increasing the num-ber of sweeps decreases the level of the background EEG noise. The left graph also shows how the noise levelsdecrease with decreasing stimulus intensity. The right graph shows how the noise levels were larger in the twogroups of elderly subjects than in the young normal subjects.

Table 4. Variability of Estimated Thresholds (dB SL)

Group Frequency 6 Sweeps 36 Sweeps Probability*(Hz) Mean SD Mean SD F-test Levene

Young 500 34.8 15.9 21.3 8.3 0.002 0.04178–95 Hz 1000 16.3 8.3 6.8 7.8 0.001 ns

2000 18.5 8.9 9.5 5.8 0.001 0.0824000 23.0 15.2 13.0 6.8 0.010 0.004

Young 500 36.3 13.7 25.8 11.2 0.011 ns108–125 Hz 1000 22.8 13.2 13.8 10.2 0.021 ns

2000 23.0 13.9 11.0 6.2 0.001 0.0014000 22.0 10.4 13.5 5.2 0.002 0.026

Elderly 500 39.8 16.8 21.3 14.0 0.001 ns78–95 Hz 1000 28.8 16.8 7.3 10.2 <0.001 0.003

2000 28.8 14.2 11.3 7.0 <0.001 0.0204000 33.0 18.7 11.5 9.6 <0.001 0.004

Hearing 500 25.3 19.8 10.8 18.4 0.021 nsImpaired 1000 8.5 17.6 -4.0 9.0 0.007 0.03978–95 Hz 2000 16.0 14.6 2.5 11.3 0.002 ns

4000 15.3 16.7 5.3 12.3 0.037 ns

* Probability of the null hypothesis. The F-test evaluates the difference between the means, and the Levene test evaluatesthe difference between the variances. Only values below 0.10 are tabulated.

were closer to behavioral thresholds for thehearing-impaired subjects. Both thedifferences between the physiological andthe behavioral thresholds and the variabilityof this measurement decreased as theduration of the recording increased. We shalldiscuss these findings in the context ofobjective audiometry.

Accurately estimating behavioralthresholds from physiological measurementsis the goal of objective audiometry. The usualapproach is to measure the lowest intensityat which a physiological response can bedetected and then to use this physiologicalthreshold to estimate the behavioralthreshold. The estimation can involve thesimple subtraction of the normal meandifference between physiological andbehavioral thresholds, or a more complicatedregression of the physiological threshold tothe behavioral threshold, that is, subtractinga value that varies with the intensity level ofthe physiological threshold (Rance et al, 1995;Rance and Briggs, 2002).

In both cases, the accuracy of estimating

behavioral thresholds depends upon thevariability of the physiological threshold fromone subject to the next (or from one recordingto the next). The mean difference (providedthis is not too large) can be subtracted away,but one is left with the variability. Ameasurement of threshold that averages at0 dB but has a standard deviation of 20 dBis much less accurate than a measurementthat averages at 10 dB but has a standarddeviation of 10 dB. The first technique willgive individual thresholds that vary from 40dB below to 40 dB above (the range of ±2standard deviations that normally accountsfor 95% of the values), whereas the secondtechnique, once the mean is subtracted, givesindividual thresholds that vary from 20 dBbelow to 20 dB above. The intersubjectvariability in threshold estimation dependsupon the amplitude and variability of thenear-threshold ASSR and the amplitude andvariability of the residual EEG noise level ofthe recording. The noise level of the recordingwill depend upon the baseline level of EEGand muscle noise in the recording and the


150

Figure 8. Estimation of behavioral thresholds. These graphs show how increasing the number of sweepsdecreases the difference between the physiological threshold and the behavioral threshold, that is, the physi-ological thresholds expressed in dB SL. The upper right graph shows the mean data across all frequencies forthe three subject groups. Included in these data are the results for the young subjects using rates of 108–125Hz. The two sets of data for the hearing-impaired subjects are derived from 10 dB steps starting at 70 dB or65 dB.

time spent in reducing that noise (byaveraging or by increasing the duration overwhich the frequency transform is calculated).

Using some suprathreshold measurementto estimate thresholds would have theadvantage that physiological responses atthese levels are larger and more reliablymeasured. However, the estimation ofthresholds from suprathreshold data isconfounded by the possibility of recruitment.For example, using the slope of the amplitude-intensity function to extrapolate back tothreshold requires that the slope bepredictable at all levels of intensity andthreshold. In a sensorineural hearingimpairment, the effect of intensity on theamplitude of a physiological response is oftennonlinear. Both the amplitudes and the slopesof the amplitude-intensity function can benormal at intensities more than 10 dB abovethe elevated threshold.

Signals in Noise

Whether or not a response can berecognized depends upon its amplitude andthe amplitude of the background EEG noise.The amplitude of a response generallyincreases with increasing intensity (seeFigure 3). In subjects with hearingimpairment, the response is not present atintensities below threshold but increasesmore rapidly than normal above thresholduntil it reaches normal levels (diagrams in theleft column of Figure 9, deriving from themean data shown in Figure 6). Dimitrijevicet al (2002) also showed that the ASSRamplitudes were larger in the hearingimpaired than in normal subjects for stimuliof equivalent SL. We shall refer to thisphenomenon as “physiological recruitment.”The term “recruitment” originally referred tothe abnormally rapid growth in perceivedloudness rather than the change in theamplitude of a physiological response.Physiological recruitment is likely associatedwith the cerebral processes that lead toperceptual recruitment. The amplitude ofthe ASSR at a given intensity varies fromsubject to subject. The recorded amplitudedepends on many parameters, the mostimportant of which are the amount ofsynchronized current in the generators, theorientation of these generators relative tothe recording electrodes, and the impedanceof the volume conductor (brain, skull, skin).

At stimulus rates of 78–95 Hz, the amplitudescan span a range of about nine-fold, forexample, from 30 to 270 nV at 70 dB SPL innormal young subjects. This variability isgreater at 500 Hz than at other frequencies.This variability makes it possible thatresponses will not be recognized in somesubjects even at 70 dB SPL unless the noiselevels are reduced significantly below 30 nV.The variability is diagrammatically plottedin Figure 9 using approximately ±1 standarddeviations.

The residual EEG noise level in arecording also varies from subject to subject.This variability is for the most partindependent of the variability of the ASSRamplitude. It varies mainly with the level ofmuscle activity in the recording, being muchlower in relaxed or sleeping subjects than intense subjects. We found that the noise levelswere significantly higher in older subjects.This is likely related to the fact that theyoung subjects were able to sleep through thetesting whereas the older subjects found itdifficult to sleep. In addition, we found thatthe noise level decreased with decreasingstimulus intensity. This may have been inpart related to subjects being less able tosleep when the stimuli were louder. Inaddition, since our protocol started out athigh intensities and then decreased theintensity to determine threshold, lower-intensity stimuli would occur later in therecording session as the subject became morerelaxed or fell asleep. The amount of noise inthe recorded spectrum decreases withincreasing frequency (John and Picton, 2000;Picton et al, 2003). The way we presented ourstimuli (Table 2) therefore caused the noiseto be lower for stimuli with higher carrierfrequencies and for left-ear compared to right-ear stimuli.

The main determinant of the residualEEG-noise level, however, is the duration ofthe recording. The amplitude of the residualnoise decreases by the square root of thenumber of sweeps (Figure 7). A longer timespent in the recording will give lessbackground EEG noise. This will lead toeasier recognition of the responses andgreater precision in the threshold estimation.It will also cause less absolute variability inthe background noise levels, which will reducethe variability in threshold estimation. Lutset al (2004) have recently stressed the effectsof recording time on the estimation of


151

thresholds in young babies. Our measurement of the amplitude of

the noise is the root-mean-square value of theactivity in 120 bins adjacent to the signalbin. We based the Fourier transform on a16.4 second sweep, which resulted in binwidths of 0.061 Hz. If the sweep had beenshorter, the bin width would have beenbroader, and the noise amplitude measuredwithin the bin would have been higherHowever, if the shorter sweep was averagedmore (i.e., the time for the recording was thesame) prior to the transform, the noise wouldbe reduced in a way similar to increasingthe sweep duration (see John et al, 1998, forfurther discussion).

The measured amplitude of the responseis affected by the level of residual EEG noisein the recording. This noise can either add toor subtract from the actual amplitude, but theoverall tendency is to cause a largermeasurement. The Appendix (together withFigure 10) demonstrates how the amplitudeis overestimated by a factor that varies withthe ratio of the measured amplitude to thenoise. This effect is evident in the resultsshown in Table 3. As the number of sweepsincreases from 6 to 36 the average noise

levels of the recording decreases by a factorapproximately equal to the square root of 6(2.45). For example, the noise level in theresponse of the hearing-impaired subjectsdecreases from 33.3 to 12.6 nV (an actualfactor of 2.64). As the noise level decreases,the measured amplitude also decreases by asmall amount. To continue with the example,the amplitude decreases from 57.3 to 53.8 nV.If we use the calculations described in theAppendix, an unbiased estimate of theamplitude would be 51.3 nV after an averageof 6 sweeps and 52.5 nV after 36 sweeps.The amplitudes reported in this paper havenot been corrected using these procedures.Such corrections would have made the slopesof the amplitude-intensity functions a littlesteeper at the lower intensities.

The ASSRs recorded at rates of 108–125Hz in the young subjects were about one-half the size of the responses recorded atrates of 78–95 Hz. However, since the residualEEG noise at these frequencies was alsolower, the responses could still be recognizedat intensities near threshold. The variabilityand the level of the threshold estimates wereslightly but not significantly higher at thehigher rates. We therefore decided not to


152

Figure 9. Factors affecting the estimation of behavioral thresholds. The graphs on the left show diagrammaticallythe mean amplitude-intensity functions for subjects with normal hearing and with hearing impairment,together with approximate ±1 SD lines. These amplitudes are plotted in relation to normal thresholds (HL).In the graphs in the middle and on the right, the same data are plotted in relation to the actual thresholds ofthe subjects (SL). Superimposed on the amplitude-intensity functions are the noise levels that might beexpected after averaging for 6 sweeps or for 36 sweeps. These values are the mean noise levels ±1 SD, multi-plied by 1.7 to give the level of response amplitude that would be recognized as significant at p < 0.05. The aver-age threshold estimations are then indicated with the arrows.

pursue these responses in the elderly subjects.Another factor was our recent finding that thedecline in the response at higher frequenciesis greater in elderly than in young subjects(Purcell et al, 2004).

Threshold Estimation

The main goal of objective audiometry isto estimate frequency-specific thresholds inan individual subject. The accuracy of thisestimation is measured by the range of valuesobtained in a group of similar subjects.Provided that the mean difference betweenthe physiological and behavioral thresholdsis not too large, that is, less than 30 dB, it canbe subtracted away from the physiologicalthreshold to give an estimated behavioralthreshold. If the difference is large, errorsstart to occur because the expected range ofhearing loss cannot be fully covered. Also, thevariability of threshold estimation oftenvaries with the mean size of the differencebetween physiological and behavioralthresholds. Once the mean difference issubtracted, the variability of the differencebecomes the variability of thresholdestimation. The two main determinants ofthis variability are the variability in theamplitude of the response near thresholdand the variability of the noise level. This isillustrated in Figure 9.

Physiological thresholds are measuredcloser to behavioral thresholds in subjectswith sensorineural hearing impairment thanin normal subjects (Figure 8 and in Table 4).After 6 sweeps we found the averagedifference to be 23.2 and 27.6 dB in the youngand elderly subjects with normal hearing,respectively, and 16.3 dB in the elderlyhearing impaired. After 36 sweeps, the valueswere 12.6 ± 8.7, 12.4 ± 11.9 dB and 3.6 ± 13.5dB. The values obtained after 36 sweeps areof the same order as those reported by others:Dimitirijevic et al (2002) reported an averagedifference of 9.0 dB in normal subjects and7.8 dB in the hearing impaired (averageddata from their Table 3); Herdman andStapells (2001; 2003) reported averagedifferences of 11.5 and 8.8 dB. The differencesin these papers are less than reported here,perhaps because their responses wererecorded over longer periods (discussed in asubsequent paragraph). Perez-Abalo et al(2001) recording over a shorter period (up to4.5 minutes per intensity) found average

differences of 12.0 and 7.5 dB for the normaland the hearing-impaired subjects.

The likely explanation for thesedifferences is physiological recruitmentoccurring in patients with sensorineuralhearing loss. This can be seen in the diagramsin the middle and right columns of Figure 9.The amplitude-intensity functions, whichwere plotted in dB HL in the left column, arenow plotted in dB SL. The approximate levelof noise after recording for 6 or 36 sweeps isthen plotted together with its standarddeviation to show the intersubject variability.The average noise levels have been multipliedby 1.73 to give the p < 0.05 upper limits of thenoise beyond which a response would beconsidered significant (this is the square rootof the F value for degrees of freedom 2 andinfinity). The point where the mean level ofthis noise crosses the mean level of theamplitude-intensity function (indicated witharrows) gives the mean estimated threshold.The threshold in dB SL is lower for thehearing-impaired subjects.

The Melbourne group has providedextensive data showing that the regressionof the behavioral threshold on thephysiological threshold has a slope of greaterthan one (Rance et al, 1995; Cone-Wesson,2002a; Rance and Briggs, 2002; Rance andRickards, 2002). (Note that some of the earlierformulations plot physiological rather thanthe behavioral threshold on the y-axis and theslope is thus less than 1 rather than greaterthan 1). The physiological thresholds areclose to the same as the behavioral thresholdsnear 90 dB and about 30 dB higher when thebehavioral thresholds are near 0 dB HLgiving regression slopes near 1.3.

The regression slopes will, however,depend on the amount of time spent in therecording. If a longer time is spent, thephysiological thresholds become closer to thebehavioral thresholds and this change isgreater for normal-hearing subjects than forhearing-impaired subjects (see Figure 8 andTable 4). These effects are showndiagrammatically in the graphs of the middleand right columns of Figure 9. Some ballparkeffects on the regression can also be derivedin the data of Table 4. Given the averagebehavioral thresholds in our hearing-impaired subjects and normal elderly subjectsof 51 and 15 dB, the slopes of a regression forour data can be roughly estimated as thedifference between these values (36 dB)


153

divided by the difference in the estimatedthresholds—54.7 (51 plus the 3.7 dB, whichis the average of the physiological thresholdsin our hearing impaired) less 27.9 (15 plus12.9)—which gives 1.34. For the dataobtained after 6 sweeps, the approximateslope of the regression would be 1.45. Anothervariable that might affect the regressionequations is the amount of noise in therecording. We found that the noise levels innormal younger subjects were less than inolder subjects. This could move the data forthe younger subjects closer to the diagonal,making the slope closer to 1.

Regressions are typically derived fromthe threshold estimates of subjects withnormal hearing and with various degrees ofsensorineural hearing loss. The regressionsare not appropriate for patients withconductive hearing loss. These patients haveno recruitment and therefore will showsimilar amplitude-intensity functions tonormal subjects except for a shift to the right(to higher intensities). The differencesbetween physiological and behavioralthresholds will be similar to those obtainedin normal subjects.

Concluding Comments

The accuracy of threshold estimationdepends on the variability of thresholdestimation rather than on any meandifference between physiological andbehavioral thresholds. The results of theseexperiments demonstrate several ways toimprove the accuracy of estimating behavioralthresholds from the ASSRs. The main factoris the level of background EEG noise in therecording. This noise can be reduced byhaving the subject relax or fall asleep, and byspending a longer time in the recording.However, time is limited in the clinicalsituation. If we need to assess thresholdswithin about an hour (e.g., 10–15 minuterecordings at 5 or 6 intensity levels), we willtypically wind up with thresholds that havea standard deviation of about 10 dB. This issimilar to the variability obtained using toneABRs (Stapells, 2000). This means that somesubjects will have thresholds that are 20 dBhigher, and some will have thresholds that are20 dB lower (provided that the meandifference has been subtracted).

To the audiologist fitting a hearing aid,this degree of variability—far from the normal

±5 dB of the behavioral audiogram—isworrisome. It is better than no informationbut far from optimal. Several proceduresmight improve the estimation. First, we couldmeasure the size of the response at highintensities and use this to tell us what theamplitude at low intensities might be. In asubject with a small response at 80 dB HL,we might decide to average for longer periodsof time to ensure that his or her inherentlysmall response is properly recognizable atlow intensities. Second, phase-weightingmight help to facilitate the recognition of aresponse at low intensities (Picton et al,2001). Third, since reduction of thebackground EEG noise is the most importantfactor determining the threshold variability,sedation might be needed for the test. Finally,accuracy might be improved by using someBayesian approach to estimating the wholeaudiogram rather than thresholds atindividual audiometric frequencies (e.g., asused in behavioral audiometry by Eilers et al,1993).

Acknowledgment. This research was supported bythe Canadian Institutes of Health Research, theRotman Research Institute, and James Knowles.Malcom Binns provided statistical advice.

REFERENCES

Cohen LT, Rickards FW, Clark GM. (1991) A com-parison of steady-state evoked potentials to modulatedtones in awake and sleeping humans. J Acoust SocAm 90:2467–2479.

Cone-Wesson B, Dowell RC, Tomlin D, Rance G, MingWJ. (2002a) The auditory steady-state response: com-parisons with the auditory brainstem response. J AmAcad Audiol 13:173–187.

Cone-Wesson B, Parker J, Swiderski N, Rickards F.(2002b) The auditory steady-state response: full-termand premature infants. J Am Acad Audiol 13:260–269.

Cone-Wesson B, Rickards F, Poulis C, Parker J, TanL, Pollard J. (2002c) The auditory steady stateresponse: clinical observations and applications ininfants and children. J Am Acad Audiol 13:270–282.

Dimitrijevic A, John MS, Van Roon P, Purcell DW,Adamonis J, Ostroff J, Nedzelski JM, Picton TW.(2002) Estimating the audiogram using multiple audi-tory steady-state responses. J Am Acad Audiol13:205–224.

Eilers RE, Ozdamar O, Steffens ML. (1993)Classification of audiograms by sequential testing:reliability and validity of an automated behavioralhearing screening algorithm. J Am Acad Audiol4:172–181.

Herdman AT, Stapells DR. (2001) Thresholds deter-mined using the monotic and dichotic multiple


154

auditory steady-state response technique in normal-hearing subjects. Scand Audiol 30:41–49.

Herdman AT, Stapells DR. (2003) Auditory steadystate response thresholds of adults with sensorineuralhearing impairment. Int J Audiol 42:237–248

John MS, Brown DK, Muir PJ, Picton TW. (2004)Recording auditory steady-state responses in younginfants. Ear Hear 25:539–553.

John MS, Dimitrijevic A, Picton TW. (2001a) Weightedaveraging of steady-state responses. Clin Neurophysiol112:555–562.

John MS, Dimitrijevic A, van Roon P, Picton TW.(2001b) Multiple auditory steady-state responses toAM and FM stimuli. Audiol Neurootol 6:12–27.

John MS, Dimitrijevic A, Picton TW. (2002a) Auditorysteady-state responses to exponential modulationenvelopes. Ear Hear 23:106–117.

John MS, Dimitrijevic A, van Roon P, Picton TW.(2001b) Multiple auditory steady-state responses toAM and FM stimuli. Audiol Neurootol 6:12–27.

John MS, Lins OG, Boucher BL, Picton TW. (1998)Multiple auditory steady-state responses (MASTER):stimulus and recording parameters. Audiology37:59–82.

John MS, Picton TW. (2000) MASTER: a windowsprogram for recording multiple auditory steady-stateresponses. Comput Methods Programs Biomed61:125–150.

John MS, Purcell DW, Dimitrijevic A, Picton TW.(2002) Advantages and caveats when recording steady-state responses to multiple simultaneous stimuli. JAm Acad Audiol 13:246–259.

Lins OG, Picton TW. (1995) Auditory steady-stateresponses to multiple simultaneous stimuli.Electroencephalogr Clin Neurophysiol 96:420–432.

Lins OG, Picton TW, Boucher BL, Durieux-Smith A,Champagne SC, Moran LM, Perez-Abalo MC, MartinV, Savio G. (1996) Frequency-specific audiometryusing steady-state responses. Ear Hear 17:81–96.

Luts H, Desloovere C, Kumar A, Vandermeersch E,Wouters J. (2004) Objective assessment of frequency-specific hearing thresholds in babies. Int J PediatrOtorhinolaryngol 68:915–926.

Perez-Abalo MC, Savio G, Torres A, Martin V,Rodriguéz E, Galan L. (2001) Steady state responsesto multiple amplitude-modulated tones: an optimizedmethod to test frequency-specific thresholds in hear-ing-impaired children and normal-hearing subjects.Ear Hear 22:200–211.

Picton TW, Dimitrijevic A, John MS, van Roon P.(2001) The use of phase in the detection of auditorysteady-state responses. Clin Neurophysiol112:1692–1711.

Picton TW, John MS, Dimitrijevic A, Purcell DW.(2003) Human auditory steady-state responses. IntJ Audiol 42:177–219.

Picton TW, John MS, Purcell DW, Plourde G. (2003)Human auditory steady-state responses: effects ofrecording technique and state of arousal. Anesth Analg97:1396–1402.

Purcell DW, John MS, Schneider BA, Picton TW.(2004) Human temporal auditory acuity as assessedby envelope following responses. J Acoust Soc Am116:3581–3593.

Rance G, Briggs RJS. (2002) Assessment of hearingin infants with moderate to profound impairment:


155

Figure 10. Overestimation of amplitude of a two-dimensional signal measured in noise. The left side of thefigure shows a sample signal recorded in noise. The signal is represented by the arrow. The noise is representedby the circle, which plots its standard deviation. The right side of the figure graphs the amount of overestimation.The relationship can be fitted with a double exponential curve such that the overestimation factor equals 1 +0.965e-1.348X + 0.078e-0.285X where “X” is the ratio between measured signal amplitude and noise amplitude.

the Melbourne experience with auditory steady-stateevoked potential testing. Ann Otol Rhinol LaryngolSuppl 189:22–28.

Rance G, Rickards FW. (2002) Prediction of hearingthreshold in infants using auditory steady-stateresponses. J Am Acad Audiol 13:236–245.

Rance G, Rickards FW, Cohen LT, De Vidi S, ClarkGM. (1995) The automated prediction of hearingthresholds in sleeping subjects using auditory steady-state evoked potentials. Ear Hear 16:499–507.

Rickards FW, Tan LE, Cohen LT, Wilson OJ, DrewJH, Clark GM. (1994) Auditory steady-state evokedpotential in newborns. Br J Audiol 28:327–337.

Savio G, Cardenas J, Perez Abalo M, GonzalezA.,Valdes J. (2001) The low and high frequency audi-tory steady state responses mature at different rates.Audiol Neurootol 6:279–287.

Stapells DR. (2000) Threshold estimation by the tone-evoked auditory brainstem response: a literaturemeta-analysis. J Speech Lang Pathol Audiol 24:74–83.


156

APPENDIX: OVERSTIMATION OF SIGNAL AMPLITUDE

The measurement of a two-dimensional signal recorded in noise is a combination of theactual signal (arrow in the diagram at the left of Figure 10) and noise (represented by the circlesurrounding the point of the arrowhead). The noise can be considered a normally distributedtwo-dimensional variable with equal variance in both dimensions. The circle in the diagramrepresents the standard deviation. Since the distribution of the measured values is symmetricalaround the line of the signal phase, the measured phase is an unbiased estimate of the signalphase. However, the measured amplitude (distributed in and around the standard deviationcircle) more overestimates (dark shading) than underestimates (light shading) the actualamplitude. This will make the average measurements of the signal amplitude larger than theactual amplitude by a factor that varies with how large the measured amplitude is relativeto the noise. This factor was derived from stochastic modeling and plotted in the graph at theright. The range of the graph is not extended below a ratio of 1.25 since the response wouldnot be considered significant at lower levels, nor beyond 10 since the differences are then toosmall to be meaningful. The straight lines in the graph show how to calculate an unbiasedestimate of the signal amplitude. If the measured amplitude of the response is 50 nV and theresidual background EEG is 10 nV (F value of 25), an unbiased estimate of the amplitude is50 nV divided by the compensation factor for a ratio of 5:50/1.02 or 49 nV.

estimating audiometric thresholds using auditory steady ...were presented simultaneously, four to...

Documents