3D printed biomimetic cochleae and machine learning co-modelling provides clinical informatics for cochlear implant patients

Abstract

Cochlear implants (CIs) restore hearing in patients with severe to profound deafness by delivering electrical stimuli inside the cochlea. The high electrical conductivity inside the cochlea, however, causes the electrical stimulus to spread, reducing speech comprehension. Understanding the stimulus spread in an individual patient is hampered by the poor accessibility of the inner ear and by the lack of suitable in vitro or in vivo models. Here, we report on the development of a neural network model that is informed by measurements in 3D printed biomimetic cochleae with realistic conductivity and anatomy. Using four geometric features captured in patients’ pre-operative CT scans, the model can reconstruct patients’ clinical electric field imaging (EFI) profiles arising from off-stimulation positions with a 90% mean accuracy (n=6). Moreover, the model informs the stereotypical cochlear geometries prone to stimulus spread and replicates the uncommon clinical occurrence of the ‘mid-dip’ EFI profiles. This work enables on-demand printing of patient-relevant cochlear models for CI testing, and directly reveals individual patient’s in vivo cochlear tissue resistivity (0.6 – 15.9kΩcm, n=16) by CI telemetry.

3D printed biomimetic cochleae and machine learning co-modelling provides clinical informatics for cochlear implant patients

“…The analyses were performed using a Python open-source package SALib 52 . Full results of the Sobol sensitivity analysis are available from the GitHub repository 53 .…”

“…The raw data of the 3PNN validation, Sobol sensitivity analysis, the stimulus spread trend, the resistivity prediction and the uncertainty sensitivity analyses have been deposited in Github and in Zenodo under accession code 5353394 53 . Other data generated in this study are provided in the Source Data file.…”

“…The code of the neural network model and the Sobol sensitivity analysis used in this study are available on Github and through Zenodo 53 .…”