Data from past patients used to streamline adjustment of levels for cochlear implant for new patients

Cochlear implant technology gives deaf people the ability to sense sound and speech. It consists of an electrode array inserted (implanted) in the cochlea of the ear and an external device that wirelessly connects to the electrodes. Each electrode stimulates different areas of the auditory nerve based of what frequency they represent. The sensitivity of auditory nerve fibers varies from patient to patient so upper and lower limits for each electrode must be set for each patient. Currently, this is undertaken by an operator adjusting the levels while the patient gives oral feedback. This process is both time consuming and challenging in that many patients are young children who are only partly able to provide feedback. At Oslo University Hospital, data has been collected on adjustment levels and response measurements from a number of former patients (158 have been used in this project). In this paper, we consider to what extent it is possible to predict values for new patients by using various machine learning techniques on data from previous patients. Although it is not possible to achieve fully automatic adjustments, the experiments show that a good starting point can be provided for manual adjustment. Further, the work has also shown which electrodes are most important to measure to automatically predict levels of other electrodes.