Preserving Wideband Tympanometry Information With Artifact Mitigation

Abstract

Objective: Absorbance measured using wideband tympanometry (WBT) has been shown to be sensitive to changes in middle and inner ear mechanics, with potential to diagnose various mechanical ear pathologies. However, artifacts in absorbance due to measurement noise can obscure information related to pathologies and increase intermeasurement variability. Published reports frequently present absorbance that has undergone smoothing to minimize artifact; however, smoothing changes the true absorbance and can destroy important narrow-band characteristics such as peaks and notches at different frequencies. Because these characteristics can be unique to specific pathologies, preserving them is important for diagnostic purposes. Here, we identify the cause of artifacts in absorbance and develop a technique to mitigate artifacts while preserving the underlying WBT information.

Design: A newly developed Research Platform for the Interacoustics Titan device allowed us to study raw microphone recordings and corresponding absorbances obtained by WBT measurements. We investigated WBT measurements from normal hearing ears and ears with middle and inner ear pathologies for the presence of artifact and noise. Furthermore, it was used to develop an artifact mitigation procedure and to evaluate its effectiveness in mitigating artifacts without distorting the true WBT information.

Results: We observed various types of noise that can plague WBT measurements and that contribute to artifacts in computed absorbances, particularly intermittent low-frequency noise. We developed an artifact mitigation procedure that incorporates a high-pass filter and a Tukey window. This artifact mitigation resolved the artifacts from low-frequency noise while preserving characteristics in absorbance in both normal hearing ears and ears with pathology. Furthermore, the artifact mitigation reduced intermeasurement variability.

Conclusions: Unlike smoothing algorithms used in the past, our artifact mitigation specifically removes artifacts caused by noise. It does not change frequency response characteristics, such as narrow-band peaks and notches in absorbance at different frequencies that can be important for diagnosis. Also, by reducing intermeasurement variability, the artifact mitigation can improve the test-retest reliability of these measurements.

Fig. 1.

Diagram illustrating the Titan Standard System and the new Research System, which was used in this study. The two systems use the same hardware (measuring device and extension cable). The software is different, however: The Standard System uses the Titan Suite software, which makes available only processed smoothed absorbance in 1/24^th-octave bands rather than at all measured frequencies. The new Research System used in this study includes the newly available Research Platform software, which allows access to the raw and unmodified microphone recordings by enabling control of the hardware via MATLAB and allowing the user to customize measurements, signal processing and data storage. The Boys Town GUI is our customized user interface for making measurements and saving raw data and is designed using the new Research Platform.

Fig. 2.

Example WBT measurement. (A) During a WBT measurement, static ear-canal pressure varies continuously with time from +200 daPa to about –300 daPa. Plotted dots indicate the static pressures measured at each epoch (defined below). (B) During the static pressure sweep, a train of acoustic click stimuli is presented to the ear canal. Plotted in gray is the raw microphone measurement of a continuous string of acoustic click responses. For analysis, the full measurement is divided into epochs (epochs designated with upper magenta tick marks). Each epoch contains a click response, and a static ear-canal pressure is measured for each epoch (plotted in A). The shaded green epochs denote ~0 daPa and TPP.

Fig. 3.

Example of six raw absorbances computed from raw microphone measurements (without artifact mitigation) in the same ear. Measurements 1a and 1b were recorded sequentially with the same ear tip insertion. After re-insertion of ear tip, 2a and 2b were measured. Recordings 3a and 3b were after another ear tip re-insertion. Inter-measurement variations in absorbances were seen: while 1b and 3b were mostly smooth, other measurements had fluctuations across frequency; also, 1a and 2a were higher below 2 kHz than the other measurements. (Absorbances shown were at ~0 daPa.)

Fig. 4.

Artifact mitigation of absorbances from measurements with low frequency noise. (A) A raw microphone measurement (gray) with low-frequency noise around 25 ms, resulting in a non-zero value at the end of the epoch (different than that at the beginning). (C) Raw microphone output (gray) was non-zero at the beginning of the epoch and near zero at the end. (B, D) Absorbances (gray) computed from these raw measurements exhibit artifact ripples across frequency due to the mismatched beginning and end of the time waveforms. (A, C, E) The artifact mitigation involved applying a high-pass filter, a Tukey window with zero-padding after ~30 ms (plotted with green dotted line on right y-axis), and a delay of 1.8 ms. The click responses after artifact mitigation (orange) show decreased low-frequency noise and are zero at the beginning and end of the epoch. (B, D, F) The corresponding mitigated absorbances (orange) show reduced artifact. (E) This example click response has noise in the time period where the Tukey window has a value of 1. After mitigation (orange), this noise has been reduced by the high-pass filter but not eliminated. (F) The corresponding raw (gray) and mitigated (orange) absorbances show that the mitigation reduced artifact but did not eliminate variations at the lowest frequencies. (Static pressures: −105 daPa for A & B; −122 daPa=TPP for C & D; −12 daPa =TPP for E & F)

Fig. 5.

(A) Mitigated absorbances from the same six measurements of Fig. 2 have reduced artifact compared to the raw absorbances of Fig. 2. (Measurements were at ~0 daPa.) (B) Variation in absorbance due to inaccuracy of static pressure estimates. Absorbance at estimated ~0 daPa at 1 kHz plotted against the actual static pressure in the ear canal at the time of the measurement. Note that the ~0 daPa estimate ranged from +7 daPa (measurement 3b) to −9 daPa (measurement 1a), x-axis. The solid gray line is fitted to the points and shows that the variation in absorbance at estimated ~0 daPa is mostly due to the variation in actual static pressure.

Fig. 6.

Mitigation does not distort the absorbance in a measurement without appreciable noise. (A) The raw microphone click response (gray) has almost no noise. Mitigated click response is plotted in orange. (B) Absorbance before (gray) and after (orange) mitigation are similar. (Static pressure was ~0 daPa.)

Fig. 7.

Two consecutive measurements (M1 and M2) from an ear with ossicular discontinuity. (A) Measurement M1 had low-frequency noise resulting in the raw absorbance (gray) with considerable artifact. Absorbance after mitigation (orange) had reduced artifact. (B) Measurement M2 had no appreciable noise with almost identical raw absorbance and mitigated absorbance without artifact. (C) Mitigated absorbances from M1 in A (orange) and M2 in B (brown) are similar, demonstrating that our mitigation is effective in measurements in pathological ears with ossicular discontinuity, preserving the characteristic features and reducing inter-measurement variability. (Static pressures were at TPP for both measurements.)

Fig. 8.

Two consecutive measurements (M1 and M2) from the same ear with superior canal dehiscence. (A) Raw absorbance (gray) with artifact from measurement M1 with low-frequency noise, and mitigated absorbance (orange) with reduced artifact. (B) Measurement M2 without appreciable noise result in raw and mitigated absorbances that are almost the same. (C) Superimposed mitigated absorbances from M1 in A (orange) and M2 in B (brown) are similar, which shows that the artifact mitigation technique is also effective in measurements in pathological ears with superior canal dehiscence. (Static pressures were at TPP for both measurements.)

Fig. 9.

Burst of prolonged low-frequency noise: (A) Example microphone waveform containing low-frequency noise over several consecutive epochs (presumably physiological). (B) From the 5^th epoch (labeled in A) with this type of noise, raw absorbance (gray) and mitigated absorbance (orange) both have large artifacts. (C) The 25^th epoch for comparison has raw absorbance (gray) with some artifact, and this artifact is no longer visible in the mitigated absorbance (orange). (Static pressures were +221 daPa in B and ~0 daPa in C).

Fig. 10.

Noise of large spike: (A) Example microphone waveform containing a large spike, before (gray) and after mitigation (orange). (B) Absorbance computed from the waveform before (gray) and after mitigation (orange) are similar. (Static pressure was +197 daPa.)