gms | German Medical Science

In recent years, the resolution of cone beam computer tomography has improved such that it is possible to characterize the cochlea anatomy in-vivo before surgery as well as to estimate the electrode positions within the cochlea after surgery. Some potential applications of image analysis include the determination of electrodes outside the cochlea, the detection of electrode flip-overs and the estimation of the corresponding frequency to each electrode location. Currently the method to characterize the geometry of the cochlea as well as the electrode positions is mainly manual. The method is therefore subject to human appreciation errors and it is time consuming when a large number of subjects need to be analysed.

Even if post-operative scans are contaminated by the electrical artefact caused by the metallic parts of the cochlear implant, it is now possible to estimate the electrode positions along the cochlea. The electrode position can be used to estimate the insertion angle of the electrodes which in turn can be used to guess the tonotopic frequency corresponding to that electrode location. Moreover, it is possible to estimate the electrode position with respect to the cochleas lateral wall or even the modiolar wall, which it has been shown to be impact the spread of current in the cochlea and it is also important to estimate the auditory nerve health status. For this reason, these features from cone beam computer tomography features may become important to optimize the fitting procedure as well as the sound coding strategy of the cochlear implant speech processor.

In this work we present an image processing algorithm that automatically detects the electrode position from con beam computer tomography volumetric scans. The algorithm is based on hidden Markov models random fields (HMMRF). The algorithm has been developed using synthetic con beam computer tomography datasets. Finally, the algorithm has been validated using real data from two cochlear implant subjects. The novel algorithm has been compared in performance with state of the algorithms showing the potential advantages of the in terms of location error HMMRF approach.

This work was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany's Excellence Strategy – EXC 2177/1 - Project ID 390895286 and the DFG project Number 396932747.