Development and Evaluation of Automated Tools for Auditory-Brainstem and Middle-Auditory Evoked Potentials Waves Detection and Annotation

Abstract

Auditory evoked potentials (AEPs) are brain-derived electrical signals, following an auditory stimulus, utilised to examine any obstructions along the brain neural-pathways and to diagnose hearing impairment. The clinical evaluation of AEPs is based on the measurements of the latencies and amplitudes of waves of interest; hence, their identification is a prerequisite for AEP analysis. This process has proven to be complex, as it requires relevant clinical experience, and the existing software for this purpose has little practical use. The aim of this study was the development of two automated annotation tools for ABR (auditory brainstem response)- and AMLR (auditory middle latency response)-tests. After the acquisition of 1046 raw waveforms, appropriate pre-processing and implementation of a four-stage development process were performed, to define the appropriate logical conditions and steps for each algorithm. The tools' detection and annotation results, regarding the waves of interest, were then compared to the clinicians' manual annotation, achieving match rates of at least 93.86%, 98.51%, and 91.51% respectively, for the three ABR-waves of interest, and 93.21%, 92.25%, 83.35%, and 79.27%, respectively, for the four AMLR-waves. The application of such tools in AEP analysis is expected to assist towards an easier interpretation of these signals.

Keywords: auditory brainstem response (ABR); auditory evoked potential (AEP); auditory middle latency response (AMLR); automated wave-annotation; waveforms.

Figure 2

The development process of each automated annotation tool.

Figure 1

Typical annotated ABR signal, presenting the five waves of interest, from I to V (red waveform) and AMLR signal, presenting the four waves of interest, Na, Pa, Nb and Pb (blue waveform).

Figure 3

Visualisation of an AMLR waveform from raw data (in black).

Figure 4

Visualisation of an ABR waveform from raw data (in black).

Figure 5

Visualization of an AMLR waveform after applying the visual display filters (raw signal is depicted in black and filtered signal is depicted in dark red).

Figure 6

Visualization of an ABR waveform after applying the visual display filters (raw signal is depicted in black and filtered signal is depicted in blue).

Figure 7

Detection of all local extremes (peaks in green, troughs in red) of the filtered AMLR signal (raw signal is depicted in black and filtered signal is depicted in dark red).

Figure 8

Detection of all local extremes (peaks in green, troughs in red) of the filtered ABR signal (raw signal is depicted in black and filtered signal is depicted in blue).

Figure 9

Detection of the AMLR signal’s waves of interest as resulting from the implementation of the AMLR automated annotation tool.

Figure 10

Detection of the ABR signal’s waves of interest as resulting from the implementation of the ABR automated annotation tool.

Figure 11

A screenshot of the selected AMLR waveform from the Interacoustics Interface.

Figure 12

The flowchart of the automated waves detection algorithm for an ABR waveform.

Figure 13

The flowchart of the automated waves detection tool of an AMLR waveform.

Figure 13

The flowchart of the automated waves detection tool of an AMLR waveform.

Figure 14

An illustration of a waveform in which a PAM artefact has been recorded: (a) the scale that has been used on the y-axis for the AMLRs plotting is also applied here; (b) visualization of the same waveform with a modified scale on the y-axis, to demonstrate the entire signal (raw signal is depicted in black and filtered signal is depicted in blue).

Figure 15

Selecting the waves of interest in a waveform where it is not easy to detect them.