888 Artificial Neural Network Analysis of Evoked Auditory Brainstem Responses for Prediction of Active Auditory Brainstem Implant (ABI) Electrodes in Patients With Neurofibromatosis Type II

Abstract

INTRODUCTION: Auditory brainstem implantation (ABI) is used to provide auditory sensations through electric stimulation of the cochlear nucleus and is a key strategy for hearing rehabilitation in patients with neurofibromatosis type II (NF2). The exact position of the 3 by 7 electrode array with respect to the cochlear nucleus is not known, due to complex anatomy following tumor removal, procedures during surgery, potential distortions to the anatomy, and post-operative array migration, and other factors. METHODS: A retrospective study of a prospectively maintained database of ABI users with evoked brainstem responses was carried out at an academic institution. Multi-class classification was conducted using the multilayer perceptron (MLP) feed forward neural network model (FF-NN) to classify and predict the post-operative active electrodes. RESULTS: 1,177 electrodes were stimulated intraoperatively among 85 patients [females, n = 48 (56.6%)]. ABI was implanted to the right and left side in 44 patients (51.7%) and 43 patients (48.3%) respectively. The mean follow-up of ABI was 29.6 months (range, 18-33.4 months). Notably, the heat map showed that electrodes at the center and the medial end (E9-16) of the array showed higher likelihoods of being active, irrespective of right or left side. The network was trained for 500 epochs with a learning rate of 0.3 and achieved a mean classification of about 92.3%. The open accessible software for predicting the monopolar electrode stimulation postoperatively based on bipolar intraoperative stimulation can be found here at: https://hifla.org/abi CONCLUSIONS: An ideal MLP FF-NN structure was achieved and hence may result in understanding more precisely about the ABI stimulation and bringing consistency among the neurosurgeons and audiologists worldwide.