2019 Conference on Implantable Auditory Prostheses

14-19 July 2019

Granlibakken, Lake Tahoe

Page 35

P8: USING MACHINE LEARNING TO ESTIMATE THE FUNCTIONAL STATUS OF THE

COCHLEAR NERVE IN ADULT COCHLEAR IMPLANT USERS

Jeffrey Skidmore, Angela Pellittieri, William Jason Riggs, Shuman He

Department of Otolaryngology

–

Head and Neck Surgery, The Ohio State University, Columbus, OH, USA

Background: Cognitive impairment is highly prevalent in older CI users (e.g., Mosnier et al.,

2015, 2018). Obtaining reliable behavioral responses and subjective descriptions of how

electrical stimulation is perceived from these patients can be challenging or infeasible.

Therefore, it is desirable to develop objective tools for optimizing CI programming settings for

implanted patients with declined cognitive functions. As the initial step to achieve this long-term

goal, this study applied machine learning techniques to create an objective predictive model for

assessing the functional status of the cochlear nerve (CN) in adult CI patients. Because

neurophysiological evidence shows significantly poorer CN function in children with cochlear

nerve deficiency (CND) than in children with normal-sized CNs (He et al., 2018), results

measured in these two patient populations can serve as a perfect training set for the model. This

model created an index for the functional status of the CN based on independent variables

derived from the electrically-evoked compound action potential (eCAP) input/output (I/O)

function and the eCAP refractory recovery function.

Methods: To date, 21 children with CND, as well as 18 children and 9 adults with normal-sized

CNs, have been recruited and tested for this study. All subjects were implanted with a

Cochlear® Nucleus™ device in the test ear. For each subje

ct, the eCAP I/O function and eCAP

refractory recovery function were measured at multiple electrode locations across the electrode

array. Both functions were measured with a charge-balanced, cathodic leading, biphasic

electrical pulse. Data from children with CND and children with normal-sized CNs were used as

the training dataset for a Linear Discriminant Analysis (LDA) algorithm to find a linear predictor

function that maximized the separation in CN function between these two patient populations.

Data from the adult CI users were then used in the predictive model to generate an index for the

functional status of the CN for each adult subject.

Results: The LDA algorithm successfully created a linear function that predicts the functional

status of the CN, as evidenced by two distinct distributions of index values for children with CND

and children with normal-sized CNs. The predicted functional status of the CN for adult CI users

varied along the linear function, with index values ranging from moderate to excellent neural

function. Our results also showed that independent variables from the eCAP refractory recovery

function had more weight in the predictor model than those from the eCAP I/O function.

Conclusions: We created a preliminary model that predicts the functional status of the CN for

adult CI users. The preliminary results also suggested that CN temporal responsiveness might

be heavily dependent on its functional status.

Acknowledgments: This work was supported by the R01 grant from NIDCD/NIGMS

(1R01DC016038) and the R01 grant from NIH/NIDCD (1R01DC017846).