to the method of Escudé (2006), would result in a predictive loss of 0.6%, and predicts 77.5% of the variations in LE16:
L’E16 =-21.246 + (0.970·D1) + (0.920·D3) + (0.027·θE1) (12).
Using only one diameter as input, which may be most practical for clinical purposes in terms of making a balance between predictability and applicability, would result in a prediction of 75.7% of the variance:
L’E16 =-19.605 + (1.498·D1) + (0.027·θE1) (13).
In this respect it is important to realize that the diameter(s) in equations 11, 12 and 13 can be measured directly from the pre-operative scan without the need of special programs, while it provides a guidance for the surgeon to control the insertion depth.
As an alternative approach for development of an insertion guidance tool, the spiral fitting method was evaluated, which was an extension to fitting a logarithmic spiral to the outer wall curvature as described by Van der Marel et al. (2014). General linear regression modelling was applied to adjust the arc length, i.e., the predicted linear insertion depth (LE1), to the more medial position of the electrode array. However, using this method, the residuals between the predicted and true surgical insertion distance showed standard deviations of 1.2 mm, which are greater than the standard deviations of 0.9 mm of the final model using multiple regression analysis. In line with this, the goodness-of-fit when predicting the insertion depth was better for the regression model (mean residual error 0˚, SD 33˚) than for the spiral model (mean residual error 15˚, SD 82˚).
The poorer predictive performance of the spiral fitting method could be due to the fact that the coefficients are determined by fitting a spiral on the basis radial distances which are only measured within the basal turn of the cochlea. These coefficients are then used for predictions of the arc length from the center of the round window to a target angle, which in most cases, is located far beyond the basal turn of the cochlea. Although the method was shown to be accurate in describing the outer wall arc length in the basal turn of the cochlea, predictions of arc length far beyond this point may not be as accurate. This accuracy might be improved by adding more measurements in the basal turn and/or by extending the measuring into the second turn of the cochlea to achieve better spiral fits, which would then be based on more and wider spread measurement points. Extension of measurements into the (much smaller) second turn of the cochlea may be complicated by voxel size limitations of clinical CT-scans.
Many studies have described that the type of implant influences electrode position (Tykocinski et al. 2000; Kos et al. 2005; Radeloff et al. 2008). The fact that the present study is limited to the
highly comparable HiFocus1 and HiFocus1J electrodes of Advanced Bionics and uses a uniform surgical approach, enabled clear evaluation of the feasibility and performance of a predictive model. On the other hand, this puts also a limit to the interpretation of the data, as on the basis of the present study no assumptions can be made for other implant types or surgical techniques. Adjustments to the current model are probably necessary. These adjustments may be provided by manufacturers when introducing a new implant type on the market, as they exhaustively test new prototypes in temporal bone studies and afterwards have access to large patient populations from multiple cochlear implant centers that may use a different surgical technique.
The analysis on frequency mismatch (∆F) between mapped frequency and physiological frequency showed that there exists a relatively broad range of insertion depths for which the mismatch is 2 to 6 semitones, but ∆F was up to 23 semitones in the study sample. The clinical relevance of having a small mismatch is still under discussion in the field of cochlear implantation and this study does not provide any evidence that it results in better performance outcomes of CI patients.
It is important to note that the frequency mismatch calculations were based on the model predictions that the spiral ganglion is the target of electrical stimulation (Kalkman et al. 2014, Stakhovskaya et al. 2007). To minimize frequency mismatch for the studied HiFocus 1 or 1J array estimated insertion distance and angular insertion depth of 6.7 mm and 484˚, respectively would be required. However, if the basilar membrane (BM) is assumed to be target site of excitation a shift in these estimations is observed. In that case, the estimated optimal insertion distance and insertion depth to minimize frequency mismatch would be 8.4 mm and 537 ˚ (data not presented in the figures). So, the frequency mismatch outcomes strongly depend on assumptions about the hypothetical stimulation target.
For the studied population the mean insertion distance and insertion depth with the HiFocus 1 or 1J array were 6.5 mm and 480˚; in case of the SG as target site the average electrode array is already optimally positioned, whereas in case of the BM the position is clearly suboptimal. Moreover, it should be noted that the average achieved insertion depth with a certain electrode type also varies among different CI centers influencing the frequency mismatch (Landsberger et al. 2015).
Apart from another potential advantage of the use of insertion models in cochlear implantation surgery, it supports the need for availability and development of electrode arrays of varying lengths to accommodate different cochlear sizes. Because frequency-mismatch is a function of both insertion depth and electrode length, it is hypothesized that changing the array length adjusting tocochlear size might further minimize it.