Andrew Botros

Neural response telemetry reconsidered: I. the relevance of ecap threshold profiles and scaled profiles to cochlear implant fitting

Objective: For more than a decade, Neural Response Telemetry (NRT™) has provided direct access to the electrically evoked compound action potential (ECAP) as elicited by the Nucleus® cochlear implant. When used clinically during fitting, ECAP threshold profiles are applied by shifting the profile to the audible threshold and comfort level boundaries (the T- and C-level profiles, respectively). The resulting profiles, to date, have matched the curvature of the ECAP threshold profile exactly. When compared with psychophysical profiles, previous studies have shown that this approach incurs errors in program levels that are no better than flat or population mean profiles. However, C-level profiles are observed to be flatter than T-level profiles. Accordingly, ECAP threshold profiles are flattened in this study when applied at increasing stimulus levels, and the effectiveness of this approach is evaluated among other methods. Design: In phase I, ECAP thresholds (via AutoNRT™) and T- and C-levels were measured from 15 adult Nucleus Freedom™ implantees. Psychophysical levels were measured using pulse train stimuli at six different stimulation rates, spanning 80 to 3500 Hz. The different rates spread T- and C-levels across a range of stimulus levels. At each of these levels, a scaling factor of best fit was calculated such that the shifted ECAP threshold profile, when scaled (0 giving a flat profile, 1 giving an unmodified profile), gave the best fit to the corresponding psychophysical profile. From the 148 such T- and C-level profiles, a single profile scaling model was determined by a simple linear regression. In phase II, the model was tested on data using three separate stimulation rates (250, 900, and 2400 Hz) and 14 additional subjects. The root mean square psychophysical level mismatch of the ECAP threshold profile, the scaled ECAP threshold profile, a flat profile, and a mean population profile was calculated per subject and per stimulation rate, and the differences in the means of these calculations were compared. In phase III, 13 separate subjects evaluated the scaled ECAP-based program during a 2 wk trial, comparing the new program to a flat program and a conventional ECAP-based program with unmodified ECAP threshold profiles. A questionnaire captured their subjective preferences. Results: In phase I, the profile scaling model constructed from the data prescribed a flattening of the ECAP threshold profile with increasing mean T- or C-level (in CL units): scale = 1.38 − 0.0043 PsychoMean. In phase II, the scaled ECAP-based profiles were found to fit the psychophysical profiles significantly better in all test configurations (typically of the order of 5% dynamic range) compared with all other profiles. In phase III, 62% of subjects preferred the scaled ECAP-based program, whereas 8% preferred the conventional ECAP-based program, 15% the flat program and 15% had no preference. Analyses of the questionnaires revealed significantly higher ratings for the scaled ECAP-based programs, whereas the conventional ECAP-based programs were not rated differently than the flat programs. Conclusions: The scaled ECAP threshold profile method provides a clinically significant enhancement to ECAP-based fitting methods, confirming the value of the ECAP threshold profile to cochlear implant fitting.

Neural response telemetry reconsidered: II. The influence of neural population on the ECAP recovery function and refractoriness.

Objective: The Neural Response Telemetry (NRT™) recovery function measures the electrically evoked compound action potential (ECAP) in response to a second biphasic pulse (the probe) after masking by a first pulse (the masker). The masker-probe interval is varied and the ECAP amplitude is measured at each masker-probe interval, giving an inverse exponential recovery. The prevailing understanding of the recovery function has been that faster recovery indicates a more efficient response to the individual pulses within a pulse sequence. Psychophysical data in the past have not supported this view, and in fact, the opposite result has been observed. This study explores this phenomenon from theoretical and experimental viewpoints. Fundamentally, a distinction is made between the refractoriness of a single fiber and the refractoriness of the whole nerve. The hypothesis is that the size of the neural population heavily influences whole nerve refractoriness: large neural populations operate near threshold and are more susceptible to masking, leading to slower ECAP recovery; however, they maintain temporal responsiveness through greater numbers of nonrefractory neurons. Design: In phase I, the hearing loss durations (indicators of neural survival) of 21 adult Nucleus® Freedom™ implantees were compared with the corresponding median recovery function time-constants (calculated per implant array). The data were separated by implant (nine Contour™, 12 Straight) and the means of these two groups were compared. The Straight array, delivering broader excitation, is expected to engage a larger neural population. In phase II, a computational model of the ECAP recovery function was constructed based on data from the cat auditory nerve. The model allows the neural population size to be manipulated; accordingly, recovery functions from different neural populations were compared. In phase III, ECAP thresholds (via AutoNRT™), ECAP recovery functions, and T- and C-levels were obtained from a subset of 12 subjects. Psychophysical levels were measured using pulse train stimuli at six different stimulation rates, spanning 250 to 3500 Hz. At each electrode, the recovery function time-constant &tgr; was compared with two measures of temporal responsiveness: (i) the gradient of the linear trend in psychophysical levels with stimulation rate; and (ii) the difference between ECAP threshold (a single pulse measure) and 900 Hz T-level (a pulse train measure). Results: In phase I, a trend toward shorter recovery function time-constants with increasing hearing loss durations was observed. The mean recovery function time-constant of the Contour implant group (0.51 msec) was significantly shorter than that of the Straight implant group (0.90 msec). When, in phase II, the recovery functions from the computational model were compared at equal ECAP amplitude, the larger neural population was associated with slower ECAP recovery. In phase III, the recovery function time-constant was significantly correlated with both temporal responsiveness measures, with slower ECAP, recovery associated with greater temporal responsiveness, thus confirming the results of previous studies. Conclusions: Slower ECAP recovery, at equal loudness, is associated with larger neural populations. The collective results suggest that this neural population view of the recovery function explains the observed association between slower ECAP recovery and greater temporal responsiveness.

The next generation of Nucleus® fitting: A multiplatform approach towards universal cochlear implant management

Abstract Objective: This article provides a detailed description and evaluation of the next Nucleus® cochlear implant fitting suite. A new fitting methodology is presented that, at its simplest level, requires a single volume adjustment, and at its advanced level, provides access to 22-channel fitting. It is implemented on multiple platforms, including a mobile platform (Remote Assistant Fitting) and an accessible PC application (Nucleus Fitting Software). Additional tools for home care and surgical care are also described. Design: Two trials were conducted, comparing the fitting methodology with the existing Custom Sound™ methodology, as fitted by the recipient and by an experienced cochlear implant audiologist. Study sample:Thirty-seven subjects participated in the trials. Results: No statistically significant differences were observed between the group mean scores, whether fitted by the recipient or by an experienced audiologist. The lower bounds of the 95% confidence intervals of the differences represented clinically insignificant differences. No statistically significant differences were found in the subjective program preferences of the subjects. Conclusions: Equivalent speech perception outcomes were demonstrated when compared to current best practice. As such, the new technology has the potential to expand the capacity of audiological care without compromising efficacy.

The Virtual Flute: an advanced fingering guide generated via machine intelligence

Abstract The Virtual Flute is a web service that provides many thousands of machine-predicted alternative fingerings. Alternative fingerings can offer variations in intonation and timbre, and can be easier to play in different musical contexts. Many play multiphonics. An advanced fingering guide is invaluable when exotic effects and demanding passages are required of the player. The Virtual Flute uses an expert system that predicts musical properties from acoustic impedance spectra. Impedance spectra for the 39,744 acoustic configurations of the flute are generated by a physical model of the instrument, some parameters of which are machine-learned. We report the construction and use of The Virtual Flute (http://www.phys.unsw.edu.au/music/flute/virtual/).

Automated fingering services for woodwinds: development of a "virtual clarinet"

The Virtual Flute is a popular web service that recommends alternative fingerings for dicult passages, timbre variations, intonations or multiphonics. Its database was generated by a machine-learned expert system analysing waveguide models for all 39,744 fingerings. The relatively simple geometry of the flute and its tone holes allowed a simple yet accurate model. The development of similar systems for other woodwinds faces greater modelling and computational challenges. For example, the clarinet has a more complex geometry, with tone holes whose radius and length vary by factors of 4.2 and 2.8. Further, it has several million dierent fingerings. To achieve the required accuracy, individual measurements of each hole separately and of mouthpiece and bell, as well as several dozen fingering examples, were used to determine parameters of a still simple waveguide model. The model uses conical and cylindrical segments with parallel and shunt impedances at junctions, representing tone holes. This approach of incrementally enhancing our waveguide model allows computational advantages: an ecient, woodwind-generic software framework is built that can adapt to the instrument of interest. We report interim results with such a system, with further potential applications in the design of woodwind instruments and other acoustic duct systems.