Reducing the variability in auditory evoked response to pitch matched stimuli using Wavelet Scattering Transform

Cochlear Implant (CI) attempts to restore hearing to profoundly deaf patients by stimulating the auditory nerve using electrical signals via an electrode array implanted in the cochlear. With different cochlear size and length of the electrode array, acoustic input frequency allocated for electric stimulation often does not match the characteristic frequency of auditory nerves. Individual tuning process on frequency allocation is time consuming and typically less regarded in busy clinical practice. An objective tuning with electroencephalogram (EEG) could be a preferred alternative method. Statistically, the classical method of averaging over many trials in extracting auditory evoked response from EEG signals is highly subjected to noises. In this paper, we propose to decompose the EEG signal into their scattering coefficients using wavelet scattering transform (WST). The result shows that the scattering coefficients are able to retain the underlying physiological information even after subsequent averaging. We also demonstrated that WST is a crucial pre-process when establishing a cortical correlate to a perceptual task using EEG.