Variations in Microanatomy of the Human Modiolus: Implications for Cochlear Implants

1 Human cochlear anatomy is highly variable. The phenomenon has been first 2 described qualitatively, followed by a quantitative variability assessment with detailed 3 anatomical models of the human cochlea. However, all previous work focused on 4 lateral cochlear wall. Few information is available on the variability of the modiolar 5 wall. Modiolar variability, likely determined by variability in the spiral ganglion, 6 provides key information on when during ontogenesis the individual cochlear 7 morphology is established: before and/or after neuronal structures are formed. In the 8 present study we analyzed 108 corrosion casts, 95 clinical cone beam computer 9 tomographies and 15 µCTs of human cochleae and observed modiolar variability of 10 similar and larger extent than the lateral wall variability. Lateral wall measures 11 correlated with modiolar wall measures significantly. ~49% of the variability has a 12 common cause, very likely established already during the time when the spiral 13 ganglion is formed. Proximity of other neuronal and vascular structures, defining the 14 remaining variability in scalar spaces, are determined later in ontogenesis, when the 15 scalae are formed. The present data further allows implications for perimodiolar 16 cochlear implants and their tip fold-overs. In particular, the data demonstrate that tip 17 fold-overs of preformed implants likely result from the morphology of the modiolus 18 (with radius changing from base to apex), and that optimal cochlear implantation of 19 perimodiolar arrays cannot be guaranteed without an individualized surgical 20 technique.


Introduction 25
The shape of the human cochlea has an intriguing three-dimensional geometry that 26 is reminiscent of the shell of a nautilus which remarkably fits to a logarithmic spiral 1-27 3 . A relation of the cochlear form to an acoustic function has been proposed 4 . The 28 suggestion, however, is neither compatible with the overall size 5-10 nor the large 29 interindividual variability of the cochlear shape (analysis in 8 ). The Pietsch-data were 30 compatible with the efficient packing hypothesis 11,12 , assuming that the anatomical 31 space restriction in the temporal bone, given by the proximity of nerves, muscles and 32 vessels (embryonically forming before the cochlear spaces 13 ), affects the 33 interindividual variability in the cochlear shape. The shape was not compatible with a 34 nautilus-like logarithmic spiral, but rather fits to a more complex polynomial spiral ( 8 , 35 comp. 14 ). 36 37 Human cochlear variability is of key importance for cochlear implantation.

Data analysis 143
The mean modiolar wall helix was computed based on the µCT data. First, the 144 segmentation models of the 15 µCT datasets were averaged, yielding an average 145 representation of the human cochlea. Based on this volumetric model the mean 146 modiolar wall helix was extracted, as is depicted in Fig. 2A. The helix was then 147 parameterized according to the ABH model 35 , i.e. such that it could be scaled 148 independently in x, y and z to match individual measures of the modiolar wall 149 diameter Amod and width Bmod (cf. Fig. 1B). Additionally, we studied the impact of modiolar variability on the risk of tip fold-over. 177 In order to do so we introduced the critical radius rfold, describing the curvature of an 178 array tip small enough to enable the array to "stand up" on the modiolar wall (i.e. the 179 critical radius that allows for a 90° angle between array tip and modiolar wall, as is 180 depicted in Fig 2D; it is considered critical since an angle > 90° between array tip and 181 modiolar wall will likely result in tip fold-over). Four values are hence important for the 182 investigations described above: the distance of the electrode array to the modiolus 183 (doff, different for three different perimodiolar arrays), the minimal distance to the 184 lateral wall (dLW), the critical curvature of the preformed electrode array tip (rfold in Fig.  185 2D) and the point of release of the electrode array from the stylet. We assumed three 186 different distances from the modiolus based on three different electrode arrays (see 187 results) and compared the resulting radius of the electrode tip (rpre) with the critical 188 radius (rfold) at the given implantation angle in all 108 corrosion casts. Using the large dataset of more than 200 human cochleae obtained with different 207 methods, we first focused on measures that can be easily obtained in all these 208 approaches. Using such strategy it was possible to compare the different methods to 209 each other and by that validate them. 210 The most straightforward comparison of variability was using the measures obtained 211 at A and B axes of the cochlea in clinical CTs, µCT and corrosion casts. Comparing 212 the three methods reveals that all measures taken at the lateral wall are similar and 213 overlapping with these techniques (Fig. 3). The differences were systematic at the In order to validate if employing these offset values yields data on metric and angular 316 insertion depth which is comparable to clinical observations we additionally took 317 standard deviation data reported in the three publications on the respective 318 perimodiolar arrays into account. Using the average shape of the modiolar wall we 319 used the model to compute the metric insertion depth (EID) necessary to achieve the 320 reported average insertion angles +/-1 standard deviation of the respective electrode 321 arrays. As shown in Fig. 7, the computed EID ranges necessary to achieve the 322 clinically observed ranges of insertion angles are very similar to the ones assessed 323 within clinical data: for the Contour Advance electrode the mean implantation angle 324 of 348 ± 36° was clinically achieved with an EID of 16.6 ± 1.1mm 45 , the model 325 prediction was nearly identical -16.7 +/-1.1mm (Fig. 7ü). For the Mid Scala 326 electrode, clinical data have shown that the mean implantation angle of 398 ± 41° 327 required an EID of 19.1 ± 0.9 mm 46 and the model prediction was again nearly 328 identical -19.2 ± 1.3mm (Fig. 7).  Interestingly, the ranges for the optimal release point Istr and the ranges critical for 361 contacts with the lateral wall Icrit overlapped for doff 0.8 and 1.0 mm. This 362 demonstrates that for these distances from the modiolus there is no universally safe 363 Istr that guarantees both (i) a safe release from straightener (without tip fold-over) and 364 (ii) no risk of trauma at the lateral wall. In other words there is no "value that fits all" 365 and the surgeon's guides for release from stylet require at least different values for 366 small, mean and large cochleae. This highlights again the importance of individually 367 assessing the patient anatomy prior to implantation. 368 Next, the interrelation of EID and IA was investigated for the different values of doff. 369 The data, consistent with Fig. 6, further suggest that if an array is located closer to 370 the modiolus, shorter insertion depths are required to achieve specific insertion 371 angles (Fig. 8C). Modiolar electrodes of a certain length can thus theoretically 372 achieve higher insertion angles than lateral wall electrodes of the same length. 373 Pragmatically, these pre-curved electrodes are never inserted beyond or even up to 374 540°, which is most likely owed to the complexity of the insertion and trajectory the 375 individual corrosion cast the critical radii rfold (as defined in Fig. 2D) were determined 408 along the first two turns of the cochlea (Fig. 9). These values were highly 409 interindividually variable. Nonetheless, within the first 270° the critical radius 410 functions were rather flat, with a maximum of the mean curve of 1.37 mm. This is of 411 importance, since the release from the straightener (e.g. stylet in case of Contour 412 Advance) must take place within the first 45°-90°, but preferentially after the end of 413 the straight portion of the implant course, thus after ~ 5 mm insertion (Fig. 8B). In 414 consequence, to safely prevent tip foldover at this position, the tip of the implant after 415 release from the stylet should have a preformed radius ≥ 1.37 mm for the average 416 cochlea such that the array tip cannot fold over within the basal cochlear region. 417 However, the value of 1.37 mm is not optimal for all cochleae; to safely avoid tip 418 foldover in all cochleae, the radius should even exceed 2 mm. 419  Fig.  463 10). The diagram in Fig. 10A further shows that after about 500°, the pre-curvature 464 radius rpre is even smaller than the foldover critical radius rfold. Foldovers beyond 465 insertion angles of 500° are hence nearly inevitable with such array design. 466 In the other arrays (Fig. 10B,C) the mean optimal curvature is always above the 467 critical curvature and this danger is consequently less (N.B. this applies for the mean 468 cochleae only). This demonstrates that for assuring atraumatic insertion without the 469 risk of tip fold-over, the electrode should be designed to be located more than 0.3 470 mm away from the modiolus. The presented data provide evidence that the modiolar cochlear structures are either 495 as variable as the cochlear lateral wall or, in some measures, even more variable 496 than the lateral wall. In no case the variability of the modiolar walkl was less than that 497 of the lateral wall. The interindividual variability of the human cochlea thus extends 498 also into the modiolus that is, in contrast to the scalar spaces, primarily shaped by 499 the early-developing neural structures. 500

501
The mechanistic explanation of cochlear variability has been so far based on the 502 efficient packing hypothesis and the fact that scala vestibuli and scala tympani form 503 after the differentiation of the surrounding neuronal structures. Since the present 504 study did not assess neuronal structures directly, it cannot exclude the possibility that 505 the neuronal structures are not variable and that only the scalar spaces approach 506 them much closer in the smaller cochleae. This is, however, unlikely: the spiral 507 ganglion is located extremely close to the scala tympani, the separation being only by 508 a thin bony shell and sometimes a vessel (Fig. 9 of 48 and Fig. 6 of 49 ; see also 50 ). 509 Therefore, interindividual differences in the modiolar axes must involve variations in 510 the 3D shape of spiral ganglion. Indeed, also in a previous study metric length of the 511 first two turns of the cochlea explained 83% of the variability of spiral ganglion length 512 ( 7 , see also 51 ). Most likely, it is already early in development when this part of the 513 variability is established, before the scalar spaces appear. This suggests another an 514 inherent source of variability of the cochlear size, potentially related to the overall 515 size of the temporal bone and thus the size of the head that is additional to the 516 efficient packing. 517 518 Methodologically, when comparing the lateral wall and the modiolar wall we need to 519 consider that the borders of the lateral wall are much better defined in all imaging 520 techniques. The modiolar wall is fenestrated, and thus the border is harder to identify 521 than the lateral wall (Fig. 1). One can assume that the outcomes of modiolar 522 measurements will be more affected by measurement imprecisions (noise) than at 523 the lateral wall. This may have substantially contributed to the larger spread of the 524 data for the normalized modiolar distributions compared to lateral wall (Fig. 4). The 525 interesting finding is, however, the high correlation (r ~ 0.7) of both measures in 526 corrosion casts (with the best spatial resolution, Fig. 3A,B). This demonstrates that 527 the results in corrosion casts are not driven by measurement "noise" (that would be 528 uncorrelated), but rather by true variability behind the data. Such common factors 529 explain 49% of the variability of lateral and modiolar dimensions. Of key importance 530 is the use of several techniques: here clinical CT was much more contaminated by 531 such uncorrelated noise, and consequently the r values were smaller, ~ 0.37. 532 Interestingly, where measurements can be performed exactly, in µCT, despite few 533 data, correlation coefficients are higher than in clinical CTs (Fig. 4). show only the empty spaces and as a negative image include, particularly in the 540 modiolar measures, the soft tissue. Additionally to the imprecisions in the 541 assessment of the modiolar wall also this may further contribute to these differences. 542

Clinical implications 544
We investigated the consequence of the modiolar variability on the cochlear 545 implantation. We have focused on three arrays that cover a wide range of distances 546 from the modiolus. The present data confirm that compared to lateral wall arrays, 547 perimodiolar implants of the same length have the potential to reach deeper into the 548 cochlea. However, this includes risks in cochlear trauma and comes at a cost of a 549 complex design that currently does not allow deep implantation (see also below): 550 since the implant must be preformed, implantations require a stylet (or straightener). 551 552 Furthermore, perimodiolar arrays require a precurved geometry. A precurved 553 electrode arrays often have a constant curvature along the arrayin other words 554 they are optimally designed for one insertion position (rpre curves in Fig. 10). Before 555 (basally to) this position the curvature will be smaller than optimal and even may be 556 smaller than the critical radius (with the consequence of tip fold-over). Beyond this 557 point (apical to it) it will be too large and thus come to lie further abmodiolarly, at an The individual optimal straight insertion depth covers a range from 2 to 5 mm 564  Fig. 2) in the present study was 6.86 -568 9.37 mm. The surgeon's guide for the Contour Advance electrode informs that 569 the electrode tip is 7.6 mm from the marker for optimal insertion. This is > 0.7 570 mm more than the corresponding space in the smallest cochlea (Fig. 8 B). 571 This means that this electrode would introduce cochlear damage at the lateral 572 wall in smaller cochleae before the stylet is removed (albeit this is the case 573 only in few cochleae; see also 53  before the lateral wall of the smallest cochleae. However, the more distant the 581 electrode from the modiolus during straightener removal, the less space is 582 available (Fig. 8B). Knowledge of the size of the straight distance (Istr) and the 583 maximum length till lateral wall is touched allows for individualizing the 584 implantation procedure; however, due to resolution of clinical CTs, use of 585 cochlear models may be needed for assessing this parameter precisely 43 . 586 2) The diameter of the modiolus decreases in the apical direction. The precurved 587 diameter is dependent on the point where the release of the array from the 588 stylet takes place (Fig. 10). The deeper the implantation, the smaller the 589 diameter. At present, perimodiolar implants are mainly designed for 590 implantation into the first turn. Nonetheless, higher cochlear coverage may 591 provide more independent information channels and thus better speech 592 understanding 17,55 . Thus, perimodiolar arrays always trade optimal position 593 and risk of tip foldover. 594 The preformed implant should consider that apically the diameter of the curvature 596 must be small to adhere to the modiolus in apical portions of the cochlea. This, 597 however, may lead to tip fold-over if the release is taking place at the end of the 598 straight portion of the implantation (after 45° implantation angle, Figs. 2, 8C and 9), 599 where the critical radius is much larger than the hypothetical optimal curvature of the 600 array tip. To prevent tip fold-over in this region, the preformed radius should exceed 601 1.37 mm. This, however, is larger than e.g. the curling radius of the Contour Advance 602 electrode array 56 . The Contour Advance, likely in the intention to avoid this, has a 603 conic straight silicone tip that extends for ~ 1 mm and is not curved. This is probably 604 intended to lean on the modiolus and prevent a foldover. Nonetheless, even 605 experienced surgeons cannot prevent tip fold-over in all cochleae with this electrode 606 20,32,33 , indicating that this approach is not always successful. To optimize the implantation procedure and to exclude the risk of a tip fold-over, the 622 present days electrode designs should aim at a distance to the modiolus of >0.3 mm 623 or provide larger curvatures (>1.37 mm, best > 2 mm) after release from the 624 straightener/stylet (Fig. 10). Clinical imaging outcomes of electrode array in use 625 within the first cochlear turn show distances in the range 0.60 -1.67 mm (for 626 Cochlear 532/632 array 0.80±0.10 mm and for 512 array 0.76±0.07 mm; data from 627 58 ). Closer locations, and thus true "modiolar hugging electrodes", particularly those 628 aiming at implantations beyond 400°, require new surgical and technical approaches 629 due to the changing diameter of the modiolus. Only electrodes that are implanted 630 more laterally and subsequently approach the modiolus slowly, after the implant has 631 been placed (e.g. by the increased temperature in the inner ear in implants 632 integrating temperature-sensitive memory materials 59 ) represent a viable approach 633 for true modiolar-hugging electrodes extending beyond the first turn of the cochlea. 634 Here, however, the approach to the modiolus should start basally and continue later 635 apically to prevent that the implant is dragged out of the cochlea (which would occur 636 if the process was opposite). Such approach may, however, involve a significant 637 force on the modiolus, with associated risk of trauma. It is worth further 638 investigations, given that modiolus-hugging electrodes in the past provided such 639 excellent channel separation (in some patients) that multi-channel compressed 640 analogue stimulation (providing temporal fine structure) could be clinically used 60 . 3) Measurement noise that constitutes a part of the 51% mentioned in the limiting 690 factor above. For modiolar wall, this imprecision is larger than for the lateral 691 wall, the extent of it is, however, not clear. 692

693
These implications suggest that a correlation should be observed between head size 694 and the cochlear size that explains the inherent variability (r 2 =0.49). Unfortunately, 695 the present clinical data do not include this information and therefore it requires 696 future studies to test this hypothesis. 697 698