Automated classification of electrically-evoked compound action potentials

摘要

Electrically-evoked compound action potentials (ECAPs) is an objective measure of peripheral neural encoding of electrical stimulation delivered by cochlear implants (CIs) at the auditory nerve level. ECAPs play a key role in automated CI fitting and outcome diagnosis, as long as presence of genuine ECAP is accurately detected automatically. Combination of ECAP amplitudes and signal-to-noise ratio are shown to efficiently detect true responses, by comparing them to subjective visual expert judgments. Corresponding optimal thresholds were calculated from Receiver-Operating-Characteristic curves. This was conducted separately on three artifact rejection methods: alternate polarity, masker-probe and modified-masker-probe. This model resulted in sensitivity and specificity error of 3.3% in learning, 3.5% in testing and 5.0% in verification. It was found that the following combination of ECAP amplitude and signal-to-noise ratio would be accurate predictors: 22 μV and 1.3 dB SNR thresholds for alternate polarity, 35 μV and -0.2 dB for masker-probe and 44 μV and -0.2 dB for modified-masker-probe.