electrode position
Improving a Computational Model of a Cochlear Implant Gabrielle O’Brien1, Steven Bierer2, Eric Shea-Brown3 and Julie Arenberg Bierer2 1University
of Washington, Neuroscience, Seattle, WA, USA; 2University of Washington Speech & Hearing Sciences, Seattle, WA, USA; 3University of Washington Applied Mathematics, Seattle, WA, USA
INTRODUCTION
RESULTS
Purpose: To assess the validity of a computational model that predicts the status of the electrode-neuron interface, specifically the relationship between electrode position, neural viability and perceptual threshold. • How can we explain the incredible variety of outcomes observed in cochlear implant recipients? One major factor is the quality of the electrode-neuron interface. In a poor interface, more current is required for perceptual threshold, leading to broader neural activation and decreased number of independent perceptual channels (Bierer and Faulkner 2010). • Electrode-neuron interface depends strongly on electrode position and neural viability
ANALYSIS
In some subjects, the model predictions match the CT estimates of electrode-neuron distance. The solution space indicates the predicted threshold given a pair of neural viability and distance coordinates. For perimodiolar electrodes, a range of thresholds are possible. But the model predicts very little variability in threshold for distant electrodes. Therefore, when it is confronted with any low threshold, the fitting procedure automatically predicts it must be nearby but with poor neuronal viability. The “shape” of solution space is generally invariant to uniform shifts in neuronal sensitivity, threshold and impedance.
Focused Broad
CT Model
Left: Channel interaction caused by distant Channel A . Right: Channel interaction caused by nonviable neurons.
• If we identify poor interfaces within an array, custom stimulation strategies may improve implant user experience. For example, deactivating channels with poor interfaces may improve speech perception (Bierer et al. 2011). • We have developed a computational model of the implanted cochlea to relate electrode position, neural viability and psychophysical threshold data (Goldwyn et al. 2010). Here, we compare its predictions to CT estimates of electrode position.
In other subjects, the model predictions partially describe the CT estimates of electrode-neuron distance.
CONCLUSIONS
COMPUTATIONAL MODEL
COMPUTATIONAL MODEL
The computational model describes the cochlea as a single compartment with an infinite cylinder geometry. Current injected into Region 1 excites a number of neurons (in Region 2) determined by a global sensitivity parameter. When 100 neurons are exited, we say perceptual threshold is achieved.
• • • In others, the model predictions poorly describe the CT estimates of electrode-neuron distance everywhere. .
•
•
METHODS
•
Model parameters: Impedance of the cylindrical and external compartments in the model was predicted from electrical field imaging according to the method of Vanpoucke (2004). The same parameters are used for all modeling results.
REFERENCES
Procedure: For each subject, psychophysical thresholds for monopolar stimulation and a focused stimulation strategy were the input to the model. A nonlinear least-squares regression estimated the electrode distance to the inner wall and the neural viability at each electrode site that the model predicts would yield the pair of focused and broad thresholds. Viability between electrode sites was interpolated with a spline function.
Bierer, J. A., & Faulkner, K. F. (2010). Identifying cochlear implant channels with poor electrode-neuron interface: partial tripolar, singlechannel thresholds and psychophysical tuning curves. Ear and hearing, 31(2), 247. Bierer, J. A., Bierer, S. M., Maloff, E. S., & Lin, A. (2011). Cochlear implant channels with high thresholds degrade medial vowel perception. The Journal of the Acoustical Society of America, 129(4), 2656-2656. Goldwyn, J. H., Bierer, S. M., & Bierer, J. A. (2010). Modeling the electrode–neuron interface of cochlear implants: effects of neural survival, electrode placement, and the partial tripolar configuration. Hearing research, 268(1), 93-104. Teymouri, J., Hullar, T. E., Holden, T. A., & Chole, R. A. (2011). Verification of computed tomographic estimates of cochlear implant array position: a micro-CT and histological analysis. Otology & neurotology: official publication of the American Otological Society, American Neurotology Society [and] European Academy of Otology and Neurotology, 32(6), 980.
STATISTICAL ANALYSIS
CT study: CT scans were performed within two years of behavioral and ECAP measures. Images were analyzed by Timothy Holden using methods verified by Teymouri and colleagues (2011). Subjects: 17 adult cochlear implant listeners participated (16 unilateral, 1 bilateral for a total of 18 ears). Eight arrays were Advanced Bionics (16 electrodes): one Helix array, one Mid-scalar, and five HiFocus 1J. Ten were manufactured by Cochlear (22 electrodes), the Nucleus Contour Softip perimodiolar electrode array. Behavioral thresholds: Stimuli were biphasic pulse trains (200.4 ms; 997.9 pulses per second) presented to electrodes 2—15 using monopolar and tripolar or phased array stimulation. Participants responded using a 2AFC adaptive procedure. Step size was 2 dB decreasing to 0.5 dB after 2 reversals. Threshold (average of the final 4 reversals) converged on the 70.7% point. The procedure was repeated 4-5 times per electrode.
TEMPLATE DESIGN © 2008
www.PosterPresentations.com
Our model is an excellent predictor of electrode position for perimodiolar electrodes. At small distances, the effects of neural viability are sufficiently strong to explain the observed variability in thresholds. Because the influence of neural viability on threshold drops off quickly with electrode distance, the model cannot explain the degree of threshold variability observed in distant electrodes using only two variables. We may need to incorporate variable neuronal sensitivity (as opposed to the current global parameter), peripheral processes, or something else novel to explain the variability in distant electrodes. There are significant differences in model predictions by scala. This may have to do with distinct impedances between compartments. We observe that varying the impedance parameters can improve the quality of predictions, but the best choice of impedance is subject-specific.
ACKNOWLEDGEMENTS The farther the electrode is from the medial wall of the cochlea (as indicated by CT data), the worse the model tends to predict position. Correlation coefficient r = -0.8604.
Most electrodes are in the scala tympani, but the model disproportionately underestimates the electrode-neuron distance for electrodes in the scala media. Student’s t-test with unequal variances rejects H0 with p < 1e-5.
The model tends to underestimate the distance of the Advanced Bionics array more frequently than the Cochlear array. The AB arrays are typically farther from the inner wall than the perimodiolar Cochlear arrays. Student’s t-test with unequal variances rejects H0 with p < 1e-5.
Funding was provided by the NIH: R01-DC012142 (JAB). ADD TRAINING GRANT. The authors thank Timothy Holden for analyzing the CT data, Emily Ellis for assisting with data collection, and the participants for their time and patience.
CONTACT INFORMATION Gabrielle O’Brien: [email protected] Steven Bierer [email protected]
Julie Bierer: [email protected] Eric Shea-Brown: [email protected]
of Washington, Neuroscience, Seattle, WA, USA; 2University of Washington Speech & Hearing Sciences, Seattle, WA, USA; 3University of Washington Applied Mathematics, Seattle, WA, USA
INTRODUCTION
RESULTS
Purpose: To assess the validity of a computational model that predicts the status of the electrode-neuron interface, specifically the relationship between electrode position, neural viability and perceptual threshold. • How can we explain the incredible variety of outcomes observed in cochlear implant recipients? One major factor is the quality of the electrode-neuron interface. In a poor interface, more current is required for perceptual threshold, leading to broader neural activation and decreased number of independent perceptual channels (Bierer and Faulkner 2010). • Electrode-neuron interface depends strongly on electrode position and neural viability
ANALYSIS
In some subjects, the model predictions match the CT estimates of electrode-neuron distance. The solution space indicates the predicted threshold given a pair of neural viability and distance coordinates. For perimodiolar electrodes, a range of thresholds are possible. But the model predicts very little variability in threshold for distant electrodes. Therefore, when it is confronted with any low threshold, the fitting procedure automatically predicts it must be nearby but with poor neuronal viability. The “shape” of solution space is generally invariant to uniform shifts in neuronal sensitivity, threshold and impedance.
Focused Broad
CT Model
Left: Channel interaction caused by distant Channel A . Right: Channel interaction caused by nonviable neurons.
• If we identify poor interfaces within an array, custom stimulation strategies may improve implant user experience. For example, deactivating channels with poor interfaces may improve speech perception (Bierer et al. 2011). • We have developed a computational model of the implanted cochlea to relate electrode position, neural viability and psychophysical threshold data (Goldwyn et al. 2010). Here, we compare its predictions to CT estimates of electrode position.
In other subjects, the model predictions partially describe the CT estimates of electrode-neuron distance.
CONCLUSIONS
COMPUTATIONAL MODEL
COMPUTATIONAL MODEL
The computational model describes the cochlea as a single compartment with an infinite cylinder geometry. Current injected into Region 1 excites a number of neurons (in Region 2) determined by a global sensitivity parameter. When 100 neurons are exited, we say perceptual threshold is achieved.
• • • In others, the model predictions poorly describe the CT estimates of electrode-neuron distance everywhere. .
•
•
METHODS
•
Model parameters: Impedance of the cylindrical and external compartments in the model was predicted from electrical field imaging according to the method of Vanpoucke (2004). The same parameters are used for all modeling results.
REFERENCES
Procedure: For each subject, psychophysical thresholds for monopolar stimulation and a focused stimulation strategy were the input to the model. A nonlinear least-squares regression estimated the electrode distance to the inner wall and the neural viability at each electrode site that the model predicts would yield the pair of focused and broad thresholds. Viability between electrode sites was interpolated with a spline function.
Bierer, J. A., & Faulkner, K. F. (2010). Identifying cochlear implant channels with poor electrode-neuron interface: partial tripolar, singlechannel thresholds and psychophysical tuning curves. Ear and hearing, 31(2), 247. Bierer, J. A., Bierer, S. M., Maloff, E. S., & Lin, A. (2011). Cochlear implant channels with high thresholds degrade medial vowel perception. The Journal of the Acoustical Society of America, 129(4), 2656-2656. Goldwyn, J. H., Bierer, S. M., & Bierer, J. A. (2010). Modeling the electrode–neuron interface of cochlear implants: effects of neural survival, electrode placement, and the partial tripolar configuration. Hearing research, 268(1), 93-104. Teymouri, J., Hullar, T. E., Holden, T. A., & Chole, R. A. (2011). Verification of computed tomographic estimates of cochlear implant array position: a micro-CT and histological analysis. Otology & neurotology: official publication of the American Otological Society, American Neurotology Society [and] European Academy of Otology and Neurotology, 32(6), 980.
STATISTICAL ANALYSIS
CT study: CT scans were performed within two years of behavioral and ECAP measures. Images were analyzed by Timothy Holden using methods verified by Teymouri and colleagues (2011). Subjects: 17 adult cochlear implant listeners participated (16 unilateral, 1 bilateral for a total of 18 ears). Eight arrays were Advanced Bionics (16 electrodes): one Helix array, one Mid-scalar, and five HiFocus 1J. Ten were manufactured by Cochlear (22 electrodes), the Nucleus Contour Softip perimodiolar electrode array. Behavioral thresholds: Stimuli were biphasic pulse trains (200.4 ms; 997.9 pulses per second) presented to electrodes 2—15 using monopolar and tripolar or phased array stimulation. Participants responded using a 2AFC adaptive procedure. Step size was 2 dB decreasing to 0.5 dB after 2 reversals. Threshold (average of the final 4 reversals) converged on the 70.7% point. The procedure was repeated 4-5 times per electrode.
TEMPLATE DESIGN © 2008
www.PosterPresentations.com
Our model is an excellent predictor of electrode position for perimodiolar electrodes. At small distances, the effects of neural viability are sufficiently strong to explain the observed variability in thresholds. Because the influence of neural viability on threshold drops off quickly with electrode distance, the model cannot explain the degree of threshold variability observed in distant electrodes using only two variables. We may need to incorporate variable neuronal sensitivity (as opposed to the current global parameter), peripheral processes, or something else novel to explain the variability in distant electrodes. There are significant differences in model predictions by scala. This may have to do with distinct impedances between compartments. We observe that varying the impedance parameters can improve the quality of predictions, but the best choice of impedance is subject-specific.
ACKNOWLEDGEMENTS The farther the electrode is from the medial wall of the cochlea (as indicated by CT data), the worse the model tends to predict position. Correlation coefficient r = -0.8604.
Most electrodes are in the scala tympani, but the model disproportionately underestimates the electrode-neuron distance for electrodes in the scala media. Student’s t-test with unequal variances rejects H0 with p < 1e-5.
The model tends to underestimate the distance of the Advanced Bionics array more frequently than the Cochlear array. The AB arrays are typically farther from the inner wall than the perimodiolar Cochlear arrays. Student’s t-test with unequal variances rejects H0 with p < 1e-5.
Funding was provided by the NIH: R01-DC012142 (JAB). ADD TRAINING GRANT. The authors thank Timothy Holden for analyzing the CT data, Emily Ellis for assisting with data collection, and the participants for their time and patience.
CONTACT INFORMATION Gabrielle O’Brien: [email protected] Steven Bierer [email protected]
Julie Bierer: [email protected] Eric Shea-Brown: [email protected]
