*3.3. Intra- and Inter-Patient Analysis*

The intra-patient analysis yielded mean Dice coefficients of 94.15 ± 0.01% and 91.51 ± 0.02% for cochlea and ST, respectively, indicating high congruency. Similarly, strong correlations were also observed for surface distance metrics (Table 1). The similarity between the cochlea and ST was also high (ρ > 0.97, *p* < 0.05), wherein a strong negative correlation was observed between surface distance errors and Dice coefficients (ρ < −0.99, *p* < 0.05) (Figure 6).


**Table 1.** Intra-patient analysis of imaging and clinical parameters. Positive values in the mean column represent a larger right cochlea and vice versa. ASSD: average symmetric surface distance; HD: Hausdorff distance; CDL: cochlear duct length; LW: lateral wall.

Interestingly, inter-patient analysis also yielded high similarity indexes in terms of imaging analysis, with cochlear and ST Dice coefficients of 93.90 ± 0.05 and 91.04 ± 0.06%, respectively. Statistical analysis revealed no significant difference (*p* > 0.05), even when subdivided into groups based on cochlear size and shape (A, B, wrapping factor). However, the inter-patient variability was four times the intra-patient variability. Another interesting observation was that there was no correlation between imaging and clinical intra-patient metrics (Figure 6). The only correlations observed were with the roller coaster factor (ρ = 0.45, *p* < 0.05), B (ρ = −0.17, *p* < 0.05) and the wrapping factor (ρ = 0.14, *p* < 0.05).

Concerning the clinical metrics defining the size and shape of the cochlea, intra-patient analysis revealed a mean difference of 0.01 and 0.05 mm for dimensions A and B, respectively, suggesting that neither of the sides is generally larger than the other. By contrast, a mean absolute difference of 0.50 and 0.08 mm was observed for the same parameters. The *t*-test revealed a statistical difference (*p* < 0.05) for most of the clinical metrics, suggesting that size and shape of contralateral ears are not the same. Interestingly, the well-known size metrics A, CDL and volume did not reveal a significant difference (Figure 6).

**Figure 6.** Intra-patient population and comparative correlation plots for imaging and clinical parameters. Strong correlations (ρ > |0.50|) between parameters are represented by scatter plots with filled circles and weak correlations are shown by empty circles. P: Pearson correlation coefficient; μ: mean; σ: standard deviation; \* depicts significant relation (*p*-value < 0.05).

The difference in B showed medium correlations with differences in cochlear duct lengths (ρ = 0.28–0.54, *p* < 0.05), wrapping factor (ρ = −0.2, *p* < 0.05), volume (ρ = 0.49, *p* < 0.05) and surface area (ρ = 0.56, *p* < 0.05), whereas A only showed weak correlations with cochlear duct length (ρ = 0.12–0.15, *p* < 0.05) and the wrapping factor (ρ = −0.10, *p* < 0.05). The roller coaster factor correlated with the height of the cochlea (ρ = 0.463, *p* < 0.05), whereas the discretized duct lengths showed medium and low correlations with most metrics (ρ < 0.66, *p* < 0.05).

#### **4. Discussion**

There is a need for large, automated population studies on cochlear anatomy to improve our understanding of the structure and its implications for CI surgery. The goal of this study was to better understand the anatomy of the cochlea and its variability in size and shape, which is important for developing less traumatic electrode arrays and insertion guidance for cochlear implantation surgery. The shape and size of the cochlea can also influence the choice of cochlear implant electrode, with flexible electrode arrays being preferred for more complex cochlear shapes, whilst rigid electrodes are more suitable for cochleae with a more straightforward shape and ossifications. Knowledge of the density and location of spiral ganglion cells can help surgeons choose an electrode array that is most likely to provide good electrical contact with the spiral ganglion cells coupled with minimal frequency mismatch and therefore exhibiting the best hearing outcomes for the patient [41]. In addition, knowledge of cochlear morphology can help surgeons in identifying any abnormalities or variations in the anatomy of the cochlea that may impact on the placement or function of the cochlear implant electrode. By understanding these variations, surgeons can tailor their surgical approach to the specific needs of each patient.

Previous studies have mostly focused on the size, rather than the shape and other parameters, and have only been able to analyze a small number of temporal bones due to the time-consuming nature of manual measurements which limits the scope of the analysis. The use of automated analysis is particularly important in the context of cochlear implant surgery, as manual measurements can be time-consuming and inconsistent. For example, a recent study found that manual measurements of cochlear duct length (CDL) had a maximum absolute intra-rater difference of 3.2 mm and the intra-rater reliability between the two radiological methods used in the study was only 0.65–0.84 [42], indicating that manual measurements may not be reliable. Furthermore, manual measurements were deemed reliable only up to 720 degrees in both CT and MRI scans.

Recent advances in automated analysis tools such as CoreSlicer 2.0 (CoreSlicer, Montreal, QC, Canada), Innersight 3D (Innersight Labs, London, UK), Arterys (Arterys Inc., Redwood Shores, CA, USA), etc., have made it possible to conduct larger studies with more robust and reliable results, as demonstrated in a recent study on cardiac anatomy which showed the feasibility and reliability of using automated analysis tools for population studies [43]; thus, similar approaches can be applied to cochlear anatomy. This study analyzed a large number of clinical temporal bone CT images using Nautilus (v20220801; Oticon Medical [31]) to determine cochlear morphology and characteristics, making it more efficient and robust than manual measurements. Nautilus is a web-based image analysis tool that supports the automatic analysis of pre-operative surgical planning and post-operative assessment for cochlear implant procedures; additionally, ithas the potential to influence the intraoperative workflow in an augmented reality setup and to control insertion forces and trajectories.

The analysis showed that cochlear morphology follows a Gaussian distribution, meaning that most cochleae fall within a typical range of sizes and shapes, with relatively few individuals falling outside of this range. Multiple recent studies have drafted cochlear duct-length prediction models based mainly on dimension A or a combination and dimensions A and B [28,40,44]. Another advantage of using AI-based automatic segmentation tools is that duct lengths can be easily computed in the original image space, decreasing dependence on such mathematical models.

Cochlear dimensions A and B were observed not to be well-correlated with each other. The study also suggests that dimension B is more correlated with cochlear duct lengths, the wrapping factor and volume than dimension A, contrary to popular belief. This suggests that cochlear B may be a more important factor in determining the optimum diameter and length of the electrode array. Moreover, the correlation between cochlear dimensions and discretized duct length increases as the cochlear angle increases, further supporting this observation.

Additionally, the study found that cochleae in female populations tend to be smaller and more tightly wound around the modiolus than male cochleae, but there is a significant overlap between the two populations. There is also a need to study the inter- and intra-individual variability of cochlear anatomy, as this can impact on the reliability of population statistics and the generalizability of findings. Some studies have suggested using contralateral ear CT images when a preoperative CT image for the target ear is not available [45]. However, more research is needed to confirm this and determine the extent of the variability addressed in this study. On average, the dimensions of both ears are similar, but there are statistically significant intra-individual differences in clinically relevant dimensions. This suggests that, while the average size and shape of the cochlea may be similar between the left and right ears, there can be significant differences between the two ears of an individual. However, the results showed that inter-individual variability is four times greater than intra-individual variability, suggesting that contralateral ear CT may be used for analysis only as a last resort if preoperative imaging is not available.

The study also found that the scala tympani size varies considerably among the population, generally decreasing along the insertion depth with dimensional jumps along the trajectory (also observed in a previous study on μCT images [27]). This means that the size of the scala tympani can change significantly as the electrode array is inserted. These findings can help reduce insertion trauma and preserve residual hearing, which, in turn, may impact on the performance of the implant.

In conclusion, the results of the study suggest that certain cochlear parameters are strongly correlated and there are sex-based differences in cochlear dimensions. The results also suggest that it may be necessary to use individualized cochlear models to accurately predict surgical outcomes and optimize implant design. The implications of this research are significant for CI surgery. The size and shape of the cochlea can affect residual hearing, as well as the translocation and tip foldovers/buckling of the electrode array. The mean size and shape of the cochlea, as well as its cross-sectional analysis along the spiral, can provide important information for determining the optimum diameter and length of the electrode array, leading to better hearing outcomes for patients.

**Author Contributions:** Conceptualization, H.D., F.P., C.R., C.V. and N.G.; methodology, R.H., A.F. and J.M.; software, R.H., A.F., J.M. and Z.W.; validation, R.H. and A.F.; formal analysis, R.H., A.F. and C.K.; investigation, R.H., A.F., R.C., C.V. and N.G.; resources, R.C., H.D., F.P. and N.G.; data curation, Z.W., C.R. and C.V.; writing—original draft preparation, R.H., A.F. and C.K.; writing—review and editing, R.H., A.F., R.C., C.K., J.M., Z.W., M.H., H.D., F.P., C.R., C.V. and N.G.; visualization, R.H., A.F. and J.M.; supervision, R.C., M.H., H.D., F.P. and N.G.; project administration, R.C., H.D., F.P. and N.G.; funding acquisition, H.D., F.P. and N.G. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** The study was conducted in accordance with the data agreement, EU General Data Protection Regulation (GDPR) and local regulations on data privacy and processing that govern the use of anonymized data provided under an agreement between CHU Nice, INRIA and Oticon Medical.

**Informed Consent Statement:** Informed consent was obtained from all patients, and all experiments were performed in accordance with relevant guidelines, regulations and in accordance with the Declaration of Helsinki.

**Data Availability Statement:** The dataset analyzed within the scope of the current study cannot be made publicly available, as it has been made available to the authors under the specific authorization of CHU Nice. This authorization does not extend to the public publication and distribution of the data. Access to the Nautilus tool is, however, available upon reasonable request at raui@oticonmedical.com.

**Acknowledgments:** The authors would like to thank Stephane Hlavacek for assistance during data processing.

**Conflicts of Interest:** R.H., A.F., R.C., C.K., M.H. and F.P. are employed at Oticon Medical, France, developers of the Nautilus tool. J.M. works as a consultant for the same company. The remaining authors declare no conflict of interest.
