Sound source localization patterns and bilateral cochlear implants: Age at onset of deafness effects

Abstract

The ability to determine a sound's location is critical in everyday life. However, sound source localization is severely compromised for patients with hearing loss who receive bilateral cochlear implants (BiCIs). Several patient factors relate to poorer performance in listeners with BiCIs, associated with auditory deprivation, experience, and age. Critically, characteristic errors are made by patients with BiCIs (e.g., medial responses at lateral target locations), and the relationship between patient factors and the type of errors made by patients has seldom been investigated across individuals. In the present study, several different types of analysis were used to understand localization errors and their relationship with patient-dependent factors (selected based on their robustness of prediction). Binaural hearing experience is required for developing accurate localization skills, auditory deprivation is associated with degradation of the auditory periphery, and aging leads to poorer temporal resolution. Therefore, it was hypothesized that earlier onsets of deafness would be associated with poorer localization acuity and longer periods without BiCI stimulation or older age would lead to greater amounts of variability in localization responses. A novel machine learning approach was introduced to characterize the types of errors made by listeners with BiCIs, making them simple to interpret and generalizable to everyday experience. Sound localization performance was measured in 48 listeners with BiCIs using pink noise trains presented in free-field. Our results suggest that older age at testing and earlier onset of deafness are associated with greater average error, particularly for sound sources near the center of the head, consistent with previous research. The machine learning analysis revealed that variability of localization responses tended to be greater for individuals with earlier compared to later onsets of deafness. These results suggest that early bilateral hearing is essential for best sound source localization outcomes in listeners with BiCIs.

Fig 1. Example data from listeners with NH and BiCIs.

The x-axis shows the angle from which the target speaker presented pink noise bursts. The y-axis shows the angle from which the listener perceived the sound. The diagonal shows perfect performance. Open circles represent mean response angles and error bars represent ±1 standard deviation. (A) Data from one listener with NH. (B) Data from three listeners with BiCIs whose subject codes are given in the top-left portion of each panel. Listener IAG’s errors were driven by variability in their responses across all target speakers. Listener IAZ’s errors were driven by highly consistent biases toward the front/center speaker for lateral target speakers.

Fig 2. Relationship between patient-dependent factors.

Each row and column correspond to a different patient-dependent factor that is given along the diagonal with a histogram showing its values. Values in each cell of the upper-triangle correspond to Pearson correlation coefficients. Plots in each cell of the lower-triangle show the relationship between two factors.

Fig 3. Sound source localization for forty-eight listeners with BiCIs.

The x-axis shows the angle from which the target speaker presented pink noise bursts. The y-axis shows the angle from which the listener perceived the sound to be presented. Each panel shows a different listener whose subject codes are given in the top-left corner. Mean ±1 standard deviation for listeners who experienced an onset of deafness either ≤5 years or >5 years are shown with green △ and magenta ▢ symbols, respectively. Individual responses are shown jittered in gray in the background. Within each group, listeners are arranged by increasing age at testing, which is given in years as “Age” in years within each panel. Additionally, the RMS error over all target angles and LSI are listed in the bottom-right of each panel.

Fig 4. Relationship between RMS error over all target angles, the LSI, and patient-dependent factors.

Individuals with an onset of deafness ≤5 or >5 years are shown with green ▽ and magenta ⬠ symbols, respectively. (A) The x-axis represents age at testing in years. The y-axis represents the RMS error, where higher values indicate poorer performance. The adjusted R² was taken from a regression including factors of onset of deafness group and age at testing. The sub-panel offset to the right shows data from listeners with NH. (B) The x-axis represents the age at testing in years. The y-axis represents the LSI, where lower values indicate poorer performance. The adjusted R² was taken from a regression including factors of onset of deafness group and age at testing. The sub-panel offset to the right shows data from NH listeners. (C) The x- and y-axes represent the LSI and RMS error, respectively. The R² corresponds to the coefficient of determination taken from a correlation between overall RMS error and the LSI.

Fig 5. Relationship between logistic parameters and patient-dependent factors.

Individuals with an onset of deafness ≤5 or >5 years are shown in green ☆ and magenta ◊ symbols, respectively. The x-axis represents age at testing in years. The adjusted R² was taken from a regression including factors of onset of deafness group and age at testing. The sub-panel offset to the right shows data from NH listeners. An illustration of the localization function is shown to the right of the sub-panels for aid of interpretation. (A) The y-axis represents the range of response angles (α–β from Eq 1) of the localization function, where lower values indicate smaller ranges of responses and poorer performance. (B) The y-axis represents the slope (σ from Eq 1) of the localization function.

Fig 6. Categories of localization performance.

Plotted as in Fig 3, except that each panel corresponds to a different category of localization performance as defined by experimenters. Each row and column represent a different distribution of mean and standard deviation (SD) of response angles, respectively.

Fig 7. Examples of frequency of assignment to specific categories defined in Fig 6.

(A), (B), and (C) correspond to listeners ICT, IAG, and IAZ, respectively. The x-axis corresponds to the mean category, the y-axis corresponds to the SD category, and the z-axis corresponds to the frequency of assignment. Since this process was repeated 50 times, there were a total of 50 assignments in each subplot.

Fig 8. Histogram indicating category assignments by onset of deafness group.

The x-axis corresponds to the mean category, the y-axis corresponds to the SD category, and the z-axis corresponds to the frequency of assignment (with a total of 16 listeners in panel A and 32 listeners in panel B). (A) and (B) correspond to the ≤5 and >5 years onset of deafness groups, respectively. The range of the z-axis was scaled based upon the number of listeners in each group, with the maximum equal to half of the group size.