A Time-Saving Alternative to "Peak-Picking" Algorithms: A Gaussian Mixture Model Feature Extraction Technique for the Neurodiagnostic Auditory Brainstem Response

OBJECTIVES

The accurate and efficient analysis of neurodiagnostic auditory brainstem responses (ABR) plays a critical role in assessing auditory pathway function in human and animal research and in clinical diagnosis. Traditional analysis of the neurodiagnostic ABR analysis involves visual inspection of the waveform and manually marking peaks and troughs. Visual inspection is a tedious and time-consuming task, especially in research where there may be hundreds or thousands of waveforms to analyze. "Peak-picking" algorithms have made this task faster; however, they are prone to the same errors as visual inspection. A Gaussian mixture model-based feature extraction technique (GMM-FET) is a descriptive model of ABR morphology and an alternative to peak-picking algorithms. The GMM-FET is capable of modeling multiple waves and accounting for wave interactions, compared with other template-matching approaches that fit single waves.

DESIGN

The present study is a secondary analysis applying the GMM-FET to 321 ABRs from adult humans from 2 datasets using different stimuli and recording parameters. Goodness-of-fit of the GMM-FET to waves I and V and surrounding waves, that is, the summating potential and waves IV and VI, was assessed, and latency and amplitude estimations by the GMM-FET were compared with estimations from visual inspection.

RESULTS

The GMM-FET had a similar success rate to visual inspection in extracting peak latency and amplitude, and there was low RMS error and high intraclass correlation between the model and response waveform. Mean peak latency differences between the GMM-FET and visual inspection were small, suggesting the two methods chose the same peak in the majority of waveforms. The GMM-FET estimated wave I amplitudes within 0.12 µV of visual inspection, but estimated larger wave V amplitudes than visual inspection.

CONCLUSIONS

The results suggest the GMM-FET is an appropriate method for extracting peak latencies and amplitudes for neurodiagnostic analysis of ABR waves I and V.

Automated extraction of auditory brainstem response latencies and amplitudes by means of non-linear curve registration

BACKGROUND AND OBJECTIVES

Animal results have suggested that auditory brainstem responses (ABRs) to transient sounds presented at supra-threshold levels may be useful for measuring hearing damage that is hidden to current audiometric tests. Evaluating such ABRs requires extracting the latencies and amplitudes of relevant deflections, or "waves". Currently, this is mostly done by human observers manually picking the waves' peaks and troughs in each individual response - a process that is both time-consuming and requiring of expert experience. Here, we propose a highly automated procedure for extracting individual ABR wave latencies and amplitudes based on the well-established methodology of non-linear curve registration.

METHODS

First, the to-be-analysed individual ABRs are temporally aligned - either with one another or, if available, with a pre-existing template - by locally compressing or stretching their time axes with smooth and invertible time warping functions. Then, the individual latencies and amplitudes of relevant ABR waves are obtained by picking the latencies of the waves' peaks and troughs on the common (aligned) time axis and combining these with the individual aligned responses and inverse time warping functions.

RESULTS

Using an example ABR data set with a wide range of response latencies and signal-to-noise ratios (SNRs), we test different choices for fitting the time warping functions. We cross-validate the warping results using independent response replicates and compare automatically and manually extracted latencies and amplitudes for ABR waves I and V. Using a Bayesian approach, we show that, for the best registration condition, automatic and manual data were statistically similar.

CONCLUSIONS

Non-linear curve registration can be used to temporally align individual ABRs and extract their wave latencies and amplitudes in a way that closely matches results from manual picking.

A simple algorithm for objective threshold determination of auditory brainstem responses

The auditory brainstem response (ABR) is a sound-evoked neural response commonly used to assess auditory function in humans and laboratory animals. ABR thresholds are typically chosen by visual inspection, leaving the procedure susceptible to user bias. We sought to develop an algorithm to automate determination of ABR thresholds to eliminate such biases and to standardize approaches across investigators and laboratories. Two datasets of mouse ABR waveforms obtained from previously published studies of normal ears as well as ears with varying degrees of cochlear-based threshold elevations (Maison et al., 2013; Sergeyenko et al., 2013) were reanalyzed using an algorithm based on normalized cross-covariation of adjacent level presentations. Correlation-coefficient vs. level data for each ABR level series were fit with both a sigmoidal and two-term power function. From these fits, threshold was interpolated at different criterion values of correlation-coefficient ranging from 0 to 0.5. The criterion value of 0.35 was selected by comparing visual thresholds to computed thresholds across all frequencies tested. With such a criterion, the mean algorithm-computed thresholds were comparable to the visual thresholds noted by two independent observers for each data set. The success of the algorithm was also qualitatively assessed by comparing averaged waveforms at the thresholds determined by the two methods, and quantitatively assessed by comparing peak 1 amplitude growth functions expressed as dB re each of the two threshold measures. Application of a cross-covariance analysis to ABR waveforms can emulate visual thresholding decisions made by highly trained observers. Unlike previous applications of similar methodologies using template matching, our algorithm performs only intrinsic comparisons within ABR sets, and therefore is more robust to equipment and investigator differences in assessing waveforms, as evidenced by similar results across the two datasets.

Auditory evoked potential response and hearing loss: a review

Hypoacusis is the most prevalent sensory disability in the world and consequently, it can lead to impede speech in human beings. One best approach to tackle this issue is to conduct early and effective hearing screening test using Electroencephalogram (EEG). EEG based hearing threshold level determination is most suitable for persons who lack verbal communication and behavioral response to sound stimulation. Auditory evoked potential (AEP) is a type of EEG signal emanated from the brain scalp by an acoustical stimulus. The goal of this review is to assess the current state of knowledge in estimating the hearing threshold levels based on AEP response. AEP response reflects the auditory ability level of an individual. An intelligent hearing perception level system enables to examine and determine the functional integrity of the auditory system. Systematic evaluation of EEG based hearing perception level system predicting the hearing loss in newborns, infants and multiple handicaps will be a priority of interest for future research.

Auditory brainstem and middle latency responses recorded at fast rates with randomized stimulation

Randomized stimulation and averaging (RSA) allows auditory evoked potentials (AEPs) to be recorded at high stimulation rates. This method does not perform deconvolution and must therefore deal with interference derived from overlapping transient evoked responses. This paper analyzes the effects of this interference on auditory brainstem responses (ABRs) and middle latency responses (MLRs) recorded at rates of up to 300 and 125 Hz, respectively, with randomized stimulation sequences of a jitter both greater and shorter than the dominant period of the ABR/MLR components. Additionally, this paper presents an advanced approach for RSA [iterative-randomized stimulation and averaging (I-RSA)], which includes the removal of the interference associated with overlapping responses through an iterative process in the time domain. Experimental results show that (a) RSA can be efficiently used in the recording of AEPs when the jitter of the stimulation sequence is greater than the dominant period of the AEP components, and (b) I-RSA maintains all the advantages of RSA and is not constrained by the restriction of a minimum jitter. The significance of the results of this study is discussed.

Automatic quality assessment and peak identification of auditory brainstem responses with fitted parametric peaks

The recording of the auditory brainstem response (ABR) is used worldwide for hearing screening purposes. In this process, a precise estimation of the most relevant components is essential for an accurate interpretation of these signals. This evaluation is usually carried out subjectively by an audiologist. However, the use of automatic methods for this purpose is being encouraged nowadays in order to reduce human evaluation biases and ensure uniformity among test conditions, patients, and screening personnel. This article describes a new method that performs automatic quality assessment and identification of the peaks, the fitted parametric peaks (FPP). This method is based on the use of synthesized peaks that are adjusted to the ABR response. The FPP is validated, on one hand, by an analysis of amplitudes and latencies measured manually by an audiologist and automatically by the FPP method in ABR signals recorded at different stimulation rates; and on the other hand, contrasting the performance of the FPP method with the automatic evaluation techniques based on the correlation coefficient, FSP, and cross correlation with a predefined template waveform by comparing the automatic evaluations of the quality of these methods with subjective evaluations provided by five experienced evaluators on a set of ABR signals of different quality. The results of this study suggest (a) that the FPP method can be used to provide an accurate parameterization of the peaks in terms of amplitude, latency, and width, and (b) that the FPP remains as the method that best approaches the averaged subjective quality evaluation, as well as provides the best results in terms of sensitivity and specificity in ABR signals validation. The significance of these findings and the clinical value of the FPP method are highlighted on this paper.

An expectation-maximization algorithm based Kalman smoother approach for single-trial estimation of event-related potentials

This paper applies an expectation-maximization (EM) based Kalman smoother (KS) approach for single-trial event-related potential (ERP) estimation. Existing studies assume a Markov diffusion process for the dynamics of ERP parameters which is recursively estimated by optimal filtering approaches such as Kalman filter (KF). However, these studies only consider estimation of ERP state parameters while the model parameters are pre-specified using manual tuning, which is time-consuming for practical usage besides giving suboptimal estimates. We extend the KF approach by adding EM based maximum likelihood estimation of the model parameters to obtain more accurate ERP estimates automatically. We also introduce different model variants by allowing flexibility in the covariance structure of model noises. Optimal model selection is performed based on Akaike Information Criterion (AIC). The method is applied to estimation of chirp-evoked auditory brainstem responses (ABRs) for detection of wave V critical for assessment of hearing loss. Results shows that use of more complex covariances are better estimating inter-trial variability.

Recording of auditory brainstem response at high stimulation rates using randomized stimulation and averaging

The recording of auditory brainstem response (ABR) at high stimulation rates is of great interest in audiology. It allows a more accurate diagnosis of certain pathologies at an early stage and the study of different mechanisms of adaptation. This paper proposes a methodology, which we will refer to as randomized stimulation and averaging (RSA) that allows the recording of ABR at high stimulation rates using jittered stimuli. The proposed method has been compared with quasi-periodic sequence deconvolution (QSD) and conventional (CONV) stimulation methodologies. Experimental results show that RSA provides a quality in ABR recordings similar to that of QSD and CONV. Compared with CONV, RSA presents the advantage of being able to record ABR at rates higher than 100 Hz. Compared with QSD, the formulation of RSA is simpler and allows more flexibility on the design of the pseudorandom sequence. The feasibility of the RSA methodology is validated by an analysis of the morphology, amplitudes, and latencies of the most important waves in ABR recorded at high stimulation rates from eight normal hearing subjects.

Limitations in the rapid extraction of evoked potentials using parametric modeling

The rapid extraction of variations in evoked potentials (EPs) is of great clinical importance. Parametric modeling using autoregression with an exogenous input (ARX) and robust evoked potential estimator (REPE) are commonly used methods for extracting EPs over the conventional moving time average. However, a systematic study of the efficacy of these methods, using known synthetic EPs, has not been performed. Therefore, the current study evaluates the restrictions of these methods in the presence of known and systematic variations in EP component latency and signal-to-noise ratios (SNR). In the context of rapid extraction, variations of wave V of the auditory brainstem in response to stimulus intensity were considered. While the REPE methods were better able to recover the simulated model of the EP, morphology and the latency of the ARX-estimated EPs was a closer match to the actual EP than than that of the REPE-estimated EPs. We, therefore, concluded that ARX rapid extraction would perform better with regards to the rapid tracking of latency variations. By tracking simulated and empirically induced latency variations, we conclude that rapid EP extraction using ARX modeling is only capable of extracting latency variations of an EP in relatively high SNRs and, therefore, should be used with caution in low-noise environments. In particular, it is not a suitable method for the rapid extraction of early EP components such as the auditory brainstem potential.

Evaluation of wavelet techniques in rapid extraction of ABR variations from underlying EEG

The aim of this study is to analyse an effective wavelet method for denoising and tracking temporal variations of the auditory brainstem response (ABR). The rapid and accurate extraction of ABRs in clinical practice has numerous benefits, including reductions in clinical test times and potential long-term patient monitoring applications. One method of achieving rapid extraction is through the application of wavelet filtering which, according to earlier research, has shown potential in denoising signals with low signal-to-noise ratios. The research documented in this paper evaluates the application of three such wavelet approaches on a common set of ABR data collected from eight participants. We introduced the use of the latency-intensity curve of ABR wave V for performance evaluation of tracking temporal variations. The application of these methods to the ABR required establishing threshold functions and time windows as an integral part of the research. Results revealed that the cyclic-shift-tree-denoising performed superior compared to other tested approaches. This required an ensemble of only 32 epochs to extract a fully featured ABR compared to the 1024 epochs with conventional ABR extraction based on linear moving time averaging.

The detection of adult cortical auditory evoked potentials (CAEPs) using an automated statistic and visual detection

The detection of adult cortical auditory evoked potentials (CAEPs) can be challenging when the stimulus is just audible. The effectiveness of a statistic compared with expert examiners in (1) detecting the presence of CAEPs when stimuli were present, and (2) reporting the absence of CAEPs when no stimuli were present, was investigated. CAEPs recorded from ten adults, using two speech-based stimuli, five stimulus presentation levels, and non-stimulus conditions, were given to four experienced examiners who were asked to determine if responses to auditory stimulation could be observed, and their degree of certainty in making their decision. These recordings were also converted to multiple dependent variables and Hotelling's T2 was applied to calculate the probability that the mean value of any linear combination of these variables was significantly different from zero. Results showed that Hotelling's T2 was equally sensitive to the best of individual experienced examiners in differentiating a CAEP from random noise. It is reasonable to assume that the difference in response detection for a novice examiner and Hotelling's T2 would be even greater.

Assessment of the role of the cochlear latency effect in lateralization of click sounds in humans

Interaural time and intensity disparities (ITD and IID) are the two cues to sound lateralization. "Time-only" hypothesis claims that an IID is first converted to an interaural afferent delay (Delta t), and is then processed by the central ITD mechanism, rendering a separate IID processor unnecessary. We tested this hypothesis by assessing the contribution of the cochlear latency effect to the psychophysical ITD/IID trading ratio. Auditory brainstem responses (ABRs) were used to measure the interaural afferent delays (Delta ts) that developed with a 20/sec dichotic click train used in the trading experiment. Except for small IIDs at low loudness levels, the physiological Delta t delay produced by an IID was significantly smaller than the ITD psychophysically traded for the same IID. We concluded that the cochlear latency effect alone cannot explain the psychophysical ITD/IID trading ratios and a separate IID mechanism must be involved.

A system to generate patient-specific stimuli for use with the auditory brainstem response test

The short-pulsed "click" stimuli most commonly used to evoke an Auditory Brainstem Response (ABR) do not account for temporal shifts along the basilar membrane and do not produce ABRs of maximal amplitude. This paper describes a system that maps these temporal shifts and generates patient-specific stimuli to compensate. This is of interest both to enhance the ABR and as a potential way to map the physiology of the basilar membrane.

Auditory brainstem response classification: A hybrid model using time and frequency features

OBJECTIVE

The auditory brainstem response (ABR) is an evoked response obtained from brain electrical activity when an auditory stimulus is applied to the ear. An audiologist can determine the threshold level of hearing by applying stimuli at reducing levels of intensity, and can also diagnose various otological, audiological, and neurological abnormalities by examining the morphology of the waveform and the latencies of the individual waves. This is a subjective process requiring considerable expertise. The aim of this research was to develop software classification models to assist the audiologist with an automated detection of the ABR waveform and also to provide objectivity and consistency in this detection.

MATERIALS AND METHODS

The dataset used in this study consisted of 550 waveforms derived from tests using a range of stimulus levels applied to 85 subjects ranging in hearing ability. Each waveform had been classified by a human expert as 'response=Yes' or 'response=No'. Individual software classification models were generated using time, frequency and cross-correlation measures. Classification employed both artificial neural networks (NNs) and the C5.0 decision tree algorithm. Accuracies were validated using six-fold cross-validation, and by randomising training, validation and test datasets.

RESULTS

The result was a two stage classification process whereby strong responses were classified to an accuracy of 95.6% in the first stage. This used a ratio of post-stimulus to pre-stimulus power in the time domain, with power measures at 200, 500 and 900Hz in the frequency domain. In the second stage, outputs from time, frequency and cross-correlation classifiers were combined using the Dempster-Shafer method to produce a hybrid model with an accuracy of 85% (126 repeat waveforms).

CONCLUSION

By combining the different approaches a hybrid system has been created that emulates the approach used by an audiologist in analysing an ABR waveform. Interpretation did not rely on one particular feature but brought together power and frequency analysis as well as consistency of subaverages. This provided a system that enhanced robustness to artefacts while maintaining classification accuracy.

On the dual structure of the auditory brainstem response in dogs

OBJECTIVE

To use the over-complete discrete wavelet transform (OCDWT) to further examine the dual structure of auditory brainstem response (ABR) in the dog.

METHODS

ABR waveforms recorded from 20 adult dogs at supra-threshold (90 and 70dBnHL) and threshold (0-15dBSL) levels were decomposed using a six level OCDWT and reconstructed at individual scales (frequency ranges) A6 (0-391Hz), D6 (391-781Hz), and D5 (781-1563Hz).

RESULTS

At supra-threshold stimulus levels, the A6 scale (0-391Hz) showed a large amplitude waveform with its prominent wave corresponding in latency with ABR waves II/III; the D6 scale (391-781Hz) showed a small amplitude waveform with its first four waves corresponding in latency to ABR waves I, II/III, V, and VI; and the D5 scale (781-1563Hz) showed a large amplitude, multiple peaked waveform with its first six waves corresponding in latency to ABR waves I, II, III, IV, V, and VI. At threshold stimulus levels (0-15dBSL), the A6 scale (0-391Hz) continued to show a relatively large amplitude waveform, but both the D6 and D5 scales (391-781 and 781-1563Hz, respectively) now showed relatively small amplitude waveforms.

CONCLUSIONS

A dual structure exists within the ABR of the dog, but its relative structure changes with stimulus level.

SIGNIFICANCE

The ABR in the dog differs from that in the human both in the relative contributions made by its different frequency components, and the way these components change with stimulus level.

Automated analysis of the auditory brainstem response using derivative estimation wavelets

In this paper, we describe an algorithm that automatically detects and labels peaks I-VII of the normal, suprathreshold auditory brainstem response (ABR). The algorithm proceeds in three stages, with the option of a fourth: (1) all candidate peaks and troughs in the ABR waveform are identified using zero crossings of the first derivative, (2) peaks I-VII are identified from these candidate peaks based on their latency and morphology, (3) if required, peaks II and IV are identified as points of inflection using zero crossings of the second derivative and (4) interpeak troughs are identified before peak latencies and amplitudes are measured. The performance of the algorithm was estimated on a set of 240 normal ABR waveforms recorded using a stimulus intensity of 90 dBnHL. When compared to an expert audiologist, the algorithm correctly identified the major ABR peaks (I, III and V) in 96-98% of the waveforms and the minor ABR peaks (II, IV, VI and VII) in 45-83% of waveforms. Whilst peak II was correctly identified in only 83% and peak IV in 77% of waveforms, it was shown that 5% of the peak II identifications and 31% of the peak IV identifications came as a direct result of allowing these peaks to be found as points of inflection.

The relationship between the auditory brain-stem response and its reconstructed waveforms following discrete wavelet transformation

OBJECTIVE

To examine the relationship between the auditory brain-stem response (ABR) and its reconstructed waveforms following discrete wavelet transformation (DWT), and to comment on the resulting implications for ABR DWT time-frequency analysis.

METHODS

ABR waveforms were recorded from 120 normal hearing subjects at 90, 70, 50, 30, 10 and 0 dBnHL, decomposed using a 6 level discrete wavelet transformation (DWT), and reconstructed at individual wavelet scales (frequency ranges) A6, D6, D5 and D4. These waveforms were then compared for general correlations, and for patterns of change due to stimulus level, and subject age, gender and test ear.

RESULTS

The reconstructed ABR DWT waveforms showed 3 primary components: a large-amplitude waveform in the low-frequency A6 scale (0-266.6 Hz) with its single peak corresponding in latency with ABR waves III and V; a mid-amplitude waveform in the mid-frequency D6 scale (266.6-533.3 Hz) with its first 5 waves corresponding in latency to ABR waves I, III, V, VI and VII; and a small-amplitude, multiple-peaked waveform in the high-frequency D5 scale (533.3-1066.6 Hz) with its first 7 waves corresponding in latency to ABR waves I, II, III, IV, V, VI and VII. Comparisons between ABR waves I, III and V and their corresponding reconstructed ABR DWT waves showed strong correlations and similar, reliable, and statistically robust changes due to stimulus level and subject age, gender and test ear groupings. Limiting these findings, however, was the unexplained absence of a small number (2%, or 117/6720) of reconstructed ABR DWT waves, despite their corresponding ABR waves being present.

CONCLUSIONS

Reconstructed ABR DWT waveforms can be used as valid time-frequency representations of the normal ABR, but with some limitations. In particular, the unexplained absence of a small number of reconstructed ABR DWT waves in some subjects, probably resulting from 'shift invariance' inherent to the DWT process, needs to be addressed.

SIGNIFICANCE

This is the first report of the relationship between the ABR and its reconstructed ABR DWT waveforms in a large normative sample.

Objective detection of brainstem auditory evoked potentials with a priori information from higher presentation levels

This paper describes a brainstem auditory evoked potentials (BAEPs) detection method based on supervised pattern recognition. A previously used pattern recognition technique relying on cross-correlation with a template was modified in order to include a priori information allowing detection accuracy. Reference is made to the patient's audiogram and to the latency-intensity (LI) curve with respect to physiological mechanisms. Flexible and adaptive constraints are introduced in the optimization procedure by means of eight rules. Several data samples were used in this study. The determination of parameters was performed through 270 BAEPs from 20 subjects with normal and high audiometric thresholds and through additional BAEPs from 123 normal ears and 14 ears showing prominent wave VI BAEPs. The evaluation of the detection performance was performed in two steps: first, the sensitivity, specificity and accuracy were estimated using 283 BAEPs from 20 subjects showing normal and high audiometric thresholds and secondly, the sensitivity, specificity and accuracy of the detection and the accuracy of the response threshold were estimated using 213 BAEPs from 18 patients in clinic. Taking into account some a priori information, the accuracy in BAEPs detection was enhanced from 76 to 90%. The patient response thresholds were determined with a mean error of 5 dB and a standard deviation error of 8.3 dB. Results were obtained using experimental data; therefore, they are promising for routine use in clinic.