Human bodies were donated to the Division of Clinical and Functional Anatomy of the Innsbruck Medical University by people who had given their informed consent prior to death for the use of their bodies for scientific and educational purposes (McHanwell et al., 2008; Riederer et al., 2012). All specimens were anonymized. There was no evidence for any malformation in any human temporal bone. All procedures for cat animal tissue were approved by the BRC located at Royal Victorian Eye and Ear Hospital and Animal Research and Ethics Committee, East Melbourne, VIC, Australia. Mice breeding and care were performed at the central animal facility in Innsbruck, Austria and experiments were approved by the Austrian Ministry of Science and Research and conformed to the Austrian guidelines on animal welfare and experimentation (BMWFW-66.011/0120-WF/V/3b/2016).

The high-resolution scans of the mouse cochlea were filtered and inspected for detecting smallest resolvable image features in the inner ear. Virtual slices were compared to 1 μm thick semi-thin plastic sections stained with toluidine blue from corresponding regions of a different mouse inner ear specimen (Figure 6).

As part of a project that addressed mean shape and shape variability in soft tissues focusing on the vestibular system (Johnson Chacko et al., 2018; Johnson Chacko et al., 2018), 50 scans acquired at 15 μm voxel size were imported to Amira^® 6.2 and nerves and structures of the membranous labyrinth were manually segmented switching between the 3 orthogonal planes. Structures such as the membranous labyrinth, perilymphatic compartments of the whole inner ear, vestibular end organs, vestibulocochlear- and facial-nerve were visualized using volume and surface renderings (Figure 7). In some specimens endolymphatic duct as well as the cochlear aqueduct were traced (Figure 7H). In one big human temporal bone specimen, selected features like the auditory ossicles and membranous labyrinth were segmented and visualized in Amira^® (Figure 8).

In order to measure volume changes for different tissue preparation steps, we compared the total volume of a single specimen before and after decalcification (Figures 9A–C), and the total volume of a single decalcified specimen before and after transfer to 70% ethanol (Figures 9D–F). Pairs of image volumes were imported into Amira^® 6.2 and filtered using two subsequent bilateral filters. Pairs of image volumes were registered using the normalized mutual information in the Register Images tool, this time allowing not only rigid transformation but also isotropic and anisotropic scaling. Scaling values in the three image axes after registration were used as indicators for change in total sample volume. In addition, cross-sectional diameters of the membranous labyrinth in all semi-circular canals were measured in 4 datasets before and after transfer to 70% ethanol at 15 locations of registered image volume pairs (corresponding slices). This dual approach provided information of changes through partial dehydration to 70% ethanol in membranous structures together with the overall volume changes in the sample that is dominated by bone.

In the first staining experiment on mouse specimens, I₂KI, and PTA provided the highest contrast of soft tissue in ossified inner ears. This allows simultaneous visualization of bone and soft tissue within the same specimen acquired in one scan at 70 KeV (Figures 1, 2). The procedure is ideal for screening purposes of smaller animals (mice, rats, cats, and guinea pigs, etc.) to detect variations and anomalies of gross anatomy, bone (densities) and bigger parts of the membranous labyrinth. Volumes of different compartments can be quantified across the whole inner ear. For I₂KI, I₂E, and OsO₄, highest X-ray densities were achieved for the cochlear nerve, while PTA yielded highest X-ray density in the bone marrow (Figure 2). The generally low measured voxel intensities of OsO₄ were most likely related to problems during fixation and/or staining in this specimen, as the second OsO₄ specimen that was embedded in resin prior to scanning (Figure 1F) showed much higher staining intensities. Gastrografin^® did not add any detectable contrast to soft tissues of the inner ear (Figures 1, 2). In the second staining experiment on mouse specimens, OsO₄ staining was compared for two different fixation regimes comparing dual fixation with aldehydes and OsO₄ versus fixation solely in OsO₄. All imaged specimens provided excellent soft tissue contrast (Figures 3A–D). Still, differences could be seen between the two fixation strategies. The tectorial membrane was well visible only in OsO₄-fixed specimens (Figures 3E,G), while the preservation of the stria vascularis was superior in the aldehyde-fixed/OsO₄-post-fixed specimens (Figures 3E,F). Visibility of nerve fiber tracks was clearly superior in decalcified specimens (Figures 3G,H) compared to ossified specimens with higher contrast of nerve bundles and neurons in the OsO₄-fixed specimen (Figures 3D,H). For a distinct presentation of nerve tissue in a decalcified specimen a single OsO₄-fixation proved to be best.

Double image acquisition in the cat specimen before and after I₂KI-staining allowed for simultaneous imaging and visualization of temporal bone and soft tissue of the inner ear without a decalcification step (Figure 4). The subtraction of a bone mask from the scan of the I₂KI-stained specimen allowed visualizing structures of the inner ear including cochlea, cochlear nerve, vestibular nerve, ampulla, and semi-circular canals in 3D (Figure 4E). Still, small soft tissue features such as Reissner’s membrane and the organ of Corti were not well visible in this image acquisition mode.

Ossified aldehyde-fixed human temporal bones post-fixed in OsO₄ and immersed in PBS provided good contrast of the bigger nerves within the inner ear canal (Figure 5A), but it was not possible to trace nerve bundles in smaller bony canals in the ossified samples. The X-ray absorption of the bone was too high to distinguish OsO₄-stained myelinated nerve fiber bundles. Fibrous soft tissue gave even lower contrast and the delicate membranes of the endolymphatic compartment were hardly visible. After removal of mineral components with ETDA, OsO₄ gave excellent contrast in big human inner ears and enabled to follow nerve fiber bundles along their course (Figures 5B,E–I). Reissner’s membrane as well as the endolymph in the vestibular system was clearly delineated from the surrounding fluid spaces (Figures 5B,D–G). Fibrous connective tissue gave high contrast, as visible in the vicinity of the round window niche in Figures 5A,B. For human temporal bones the iodine contrasting techniques of ossified specimens did not provide equal results with the equipment and settings used (Figure 5C), compared to smaller animal specimens (Figure 1A). While the vestibulocochlear nerve yielded very high contrast, membranes where not clearly visible in human specimens.

In another experiment of an ossified human cochlea we tested how enhanced image resolution can help to display soft tissue post-fixed with OsO₄. 3 μm voxel size of an epoxy resin-embedded human cochlea gave good contrast of Reissner’s membrane, nerve fiber bundles, and fibrous tissue (Figure 5D). Long integration times and averaging (eight times per projection) decreased noise and enabled to trace nerve fiber bundles. The inset in Figure 5D emphasizes the delicate osseous spiral lamina that houses the peripheral nerve fiber bundles. The loss of myelination close to the habenula perforata results in lower absorption of the nerves. The outline of the organ of Corti can be distinguished and even Corti’s tunnel can be identified (Figure 5D). The featured characteristics of unmyelinated spiral ganglion somata in humans (Glueckert et al., 2005a; Potrusil et al., 2012) impedes recognition of single neuron bodies, so quantification of neurons may not be possible with this imaging of a human cochlea in toto. The influence of voxel resolution on smallest detectable image feature size in decalcified OsO₄-post-fixed specimens was evaluated in Figures 5E–G. 15 μm voxel size is suitable to identify the membranous labyrinth and bigger nerve fiber bundles as well as fibrous tissue (Figure 5E). Like in most contrast enhancement methods we tested, the stria vascularis shows higher X-ray absorption that allows identification of this three cell layer thick tissue. Increasing the voxel resolution to 10 μm reduces the field of view and hence accessible specimen size, but still enables to visualize the human inner ear in toto (Figure 5F). Nerve fiber bundles appear much clearer and smaller individual bundles may be depicted as seen in the higher magnified inset. Increasing resolution to 5.5 μm cannot cover the whole inner ear in a single scan (Figure 5G) but enables to follow the peripheral nerve fibers as they fan out towards the sensory epithelium in horizontal maximum intensity projections with a high level of detail (Figure 5H). A maximum intensity projection in the plane perpendicular to the modiolus illustrates the contribution of different nerve bundles for the innervation of the macula sacculi, individual branches of the singular nerve and peripheral as well as central cochlear nerve fibers along the tonotopic axis (Figure 5I) strikingly clear.

Maximum achievable resolution with lab-based microCT setups was tested with a sub-volume scan (interior tomography) in the decalcified, solely OsO₄-fixed mouse inner ear and compared to histological 1 μm thick semi-thin sections (Figures 6A–D). A voxel resolution of 0.5 μm (Figures 6B,D) enabled to recognize single cells. Spiral ganglion type I neurons with diameters around 10 μm can clearly be outlined as well as mesothelial cells bulging from Reissner’s membrane into the perilymphatic fluid space (Figure 6A). Even the acellular tectorial membrane presents as a structure well separated from the spiral limbus. A semi-thin section of a corresponding region of a different animal (Figure 6B) exemplifies the high level of cellular resolution possible with these scanning parameters. Nerve fibers and the stria vascularis yielded highest contrast. Latter showed some detachment, possibly as a fixation artifact because of solely OsO₄-fixation without prior aldehyde protein fixation or mechanical stress during fixative perfusion. Horizontal views demonstrated that the high level of detail enabled to recognize single outer hair cells (Figure 6D). Fluid containing Nuel’s space around the lateral cell surface of outer hair cells provided enough local contrast for individual cell counts.

OsO₄ post-fixation in combination with decalcification of mineral components seems to be the method of choice to selectively present myelinated nerve fibers and membranous structures together with some tissues that give higher contrast (e.g., sensory epithelia of the vestibular system and stria vascularis). Manual segmentation of the enveloping nerve branches allows to measure length, course, and diameter of these structures in 15 μm voxel sized datasets. We were able to segment the outline of the sensory epithelium in the vestibular end organs as distinct structures although we meet the limit of resolution of this cell layer. For the purpose of finite element simulation of current spread with electrical stimulation we introduced a gap region between the space of the sensory epithelium and peripheral nerve fiber bundles that give higher contrast (Handler et al., 2017). This allows setting conductivity parameters for this region that contains myelinated and unmyelinated nerve fibers (close to the basal pole of the hair cells) as well as loosely arranged fibrous tissue. The endolymphatic fluid spaces can easily be traced when manually outlined every fourth and fifth slice and interpolating in between. Doing so in all three axis of the coordinate system reduces stair-step segmentation artifacts (Figures 7A–D). Figure 7E shows the 3D representation of such a segmentation approach to visualize the five vestibular end organs with their apical poles of the sensory epithelium bathed in the membranous labyrinth and connected to their corresponding nerve branches. Figure 7F shows another 3D surface rendering of a specimen imaged with 15 μm voxel resolution displaying the perilymphatic space of the scala vestibuli and vestibular apparatus with vestibulocochlear nerve branches and the endolymphatic system. The course of the facial nerve was segmented in all human specimens. A volume rendering of a different specimen provided excellent views for anatomical studies (Figure 7G). Even very small canals can be traced with little effort, such as the endolymphatic duct, cochlear aqueduct, and the parallel accessory canal guiding the inferior cochlear vein that drains blood from the cochlea. A small blood vessel follows the course of the endolymphatic duct in a parallel way. Manual segmentation of all these structures turns out to be a laborious work but automatized segmentation strategies we tested were insufficient to provide reliable recognition of delicate structures. We fully segmented 50 human temporal bones for further assessments of anatomical variation (Johnson Chacko et al., 2018) and focused on the vestibular system.

To address the need for a surgical planning of prototype vestibular electrode placement, bigger specimens are needed that contain not only the inner ear but also the mastoid bone and the middle ear. We avoided decalcification of the specimens since the information of the presence of bone besides soft tissue and blood vessels may be important. Direct perfusion of inner ear fluid spaces with fixative was not possible here, so we had to rely on sufficient penetration through the Eustachian tube, the inner ear canal, and pores of the fundus region that contain small holes in the modiolar trunk for guidance of the central cochlear axon bundles. We tested I₂KI contrast enhancement but had to incubate the two specimens for 16 weeks in a low concentration I₂KI solution to get sufficient contrast of soft tissue. Voxel resolution was raised to 25 μm voxel size for a whole specimen scan (Figure 8A). Contrast of soft tissue turned out to be very good and bone distinguished by higher X-ray absorption. 3D median filtering and adaption of contrast enabled us to recognize even bigger structures of the membranous labyrinth (Figure 8A, inset). At least for some parts of the membranous labyrinth as the region close to the oval window this technique provides sufficient contrast to segment the endolymphatic compartment without any mechanical manipulation by removing the stapes or penetrating the round window. We could display the endolymphatic space in the cochlea as well as the vestibular system. Smaller structures such as the reunion duct or endolymphatic duct cannot be covered with this technique and manual segmentation is less precise and much more time consuming due to very low contrast. Taken together, this approach may provide useful information for surgical planning. For detailed information on the membranous labyrinth smaller specimen size and techniques described above are preferable.

The comparison of a single specimen before and after decalcification revealed that specimen volume was 1.62% larger after decalcification. Plausibility of registration results was carefully controlled and validated by eye. The transfer of decalcified specimens from PBS to 70% ethanol was beneficial as helped to reduce the number of air bubbles inside specimens (Figure 9) and increased the signal-to-noise ratio based on lower X-ray density of ethanol compared to PBS. The comparison of a single decalcified specimen before and after transfer to 70% ethanol showed that specimen volume was 0.55% larger after transfer to 70% ethanol. To assess whether delicate membranous structures suffered from volume changes we manually measured the cross-sectional diameter of the membranous labyrinth in the semi-circular canals in 4 specimens (15 locations per specimen). These 60 measurements confirmed results above, as the average measured increase in diameter of semi-circular canals was 0.56% after transfer from PBS to 70% ethanol. These measurements demonstrate astonishingly low tissue changes after chemical fixation and during decalcification and partial dehydration.

High X-ray attenuation by the temporal bone, which actually is the bone with highest mineral density in the human body, impedes distinct visualization of soft tissue and nerve fibers with X-ray absorption based imaging. In this work we were able to visualize soft tissues in ossified mouse inner ears very well and found I₂KI to give highest X-ray absorption compared to other contrast enhancement methods we tested (Figure 1A). The membranous labyrinth was well outlined and nerve fibers as well as the stria vascularis could easily be differentiated. Double image acquisition in a differential approach with scans taken before and after contrast enhancement and subsequent image registration and subtraction of datasets (Figure 4) added valuable information to distinguish bone from soft tissue. This procedure may be ideal for screening purposes of smaller animals (mice, rats, cats, and guinea pigs, etc.) to detect variations and/or anomalies of gross anatomy and atrophy of the vestibulocochlear nerve in the inner ear canal. Also facial nerve abnormities may be outlined. However, tracing small nerve fiber bundles within canals inside the temporal bone was not possible and the membranous labyrinth was not always clearly visible. Moreover, we faced problems of air bubbles in the aqueous environment of the iodine solution that may have a big impact on morphometric measures (Figure 9F). Recently, protocols for microscopic dual-energy imaging (microDECT) have been published (Handschuh et al., 2017) that could also yield satisfying results for inner ear specimens. Discrimination of mineralized and soft tissues based on spectral properties could be achieved even without decalcification. Human specimens stained with I₂KI (Figure 5C) likewise failed to provide sufficient contrast for delicate membranes or nerve fiber pathways through compact bone. The densely mineralized temporal bone in the bigger human specimens (hence increased voxel size and higher tube currents) impedes visualization of delicate membranous structures. This is in fact a problem of dynamic range, as the bone makes it necessary to image samples with higher energy spectra (70 kVp), while imaging of the membranous labyrinth would yield much better results at lower energies (45 kVp). Taken together, the low attenuation in thin-walled membranous soft tissues at high scanning energies and the high attenuation in the thick-walled temporal bone lead to comparatively low contrast in soft tissues and high image noise.

OsO₄-post-fixation generated only limited image contrast in scans from ossified specimens. Scanning at much higher voxel resolutions and elevated integration time can enhance image contrast (Figure 5D). We demonstrated this in an ossified and epoxy resin-embedded human temporal bone, but higher absorption of the mineral content of the temporal bones required heavy averaging to achieve a reasonable signal-to-noise ratio. Even 3 μm resolution with advanced microCT equipment did not fulfill our needs for a high-throughput screening of human inner ears. Removing the mineral components by techniques developed for transmission electron microscopy (TEM) (Glueckert et al., 2005b) guaranteed best ultrastructural preservation and met our demands for a distinct visualization of small nerve fiber bundles within small bony canals (Figures 5E–I). A good compromise for imaging the whole human inner ear in a single scan to outline relevant soft tissue structures was found at 15 μm isometric voxel resolution (Figuress 5E, 7A–D, 9E), yielding a field of view of roughly 30 mm. Imaging the same specimens at 10 μm or even 5.5 μm voxel resolution is practicable and provide fantastic details of vestibulocochlear neuronal innervation pathways (Figures 5G–I). Our 2K × 2K detector required more than one scan to cover a whole human inner ear specimen in toto due to smaller field of view size (roughly 20 mm FOV for 10 μm voxel size and 11 mm FOV for 5.5 μm voxel size) with our equipment. Larger field of view scanners with bigger detectors would solve this limitation.

In ossified specimens, OsO₄ post-fixation could not provide sufficient contrast of soft tissue. Decalcification after OsO₄-post-fixation and lower energy (45 kVp) imaging fulfilled our needs for a distinct presentation of nerve fiber bundles and the membranous labyrinth in human samples. So far, OsO₄ was mainly used for specimen preparation in electron microscopy. OsO₄ crosslinks unsaturated fatty acids covalently, which ensures fixation of the fatty cellular membranes. Formaldehyde and glutaraldehyde crosslink proteins but are not able to chemically fix fatty acids. Therefore OsO₄ cannot be attributed as a classical contrast agent that somehow attaches to soft tissue structures. It is a post-fixative that provides strong additional X-ray attenuation contrast because of the heavy metal component. The fixation effect adds another big advantage over other contrast agents. Since the amount of covalently bound OsO₄ is proportional to the amount of membranes, myelinated nerve fibers show highest contents of this heavy metal. Each myelinated central axon is about 2–3 μm in diameter surrounded by a dense arrangement of 60–83 membrane layers (Spoendlin and Schrott, 1989). The 35,000 bipolar spiral ganglion neurons in humans send out a central axon towards the brainstem and a peripheral axon to innervate the hair cells. Peripheral axons reveal only half the diameter of the central axon neurons (1–2 μm) and are ensheathed by only 20–23 myelin membranes in human (Spoendlin and Schrott, 1989). This explains the high contrast in the central nerve compared to the peripheral parts that fan out in the osseous spiral lamina and vestibular end organs. Higher incorporation of osmium ensures highest X-ray absorption in the big myelinated nerves and silhouetted them against surrounding soft tissue that contain lower membrane densities.

If information from the bone (density) and soft tissue outline is needed a dual approach with scanning the OsO₄ contrasted specimen before and after decalcification and registering both datasets can cover these needs as we showed in Figures 5A,B. In the present study, we scanned 48 specimens before and after decalcification to be able to address this aspect in a future work.

Solely fixation in OsO₄ yielded higher contrast of the myelinated nerves and is also a technique used in TEM when highest contrast of membranes is the prime goal. The sub-volume scan of the OsO₄-fixed mouse inner ear demonstrates that current lab-based microCT scanners provide sufficient image resolution for counting neurons in the spiral ganglion and even outer hair cells. The inferior fixation of proteins and hence partial removal during washing steps may emphasize cellular membranes even better and give higher contrast in microCT as well as in TEM. The main limitation of high-resolution scanning at cellular level is the comparatively long exposure time for single projection images, which could lead to a total scanning time in the range several days for one specimen.

The OsO₄-fixation approach combined with imaging before and after decalcification may be ideal for a rather fast characterization of anomalies of gross anatomy in smaller animals (mice, rats, cats, and guinea pigs, etc.) for giving quantitative information about hair cells and nerve fiber densities as well as bone density. Substantial loss of nerve tissue as well as degeneration of spiral ligament and stria vascularis should be able to be detected and volumes quantified with high precision across the whole inner ear. This non-destructive methodology covers several imaging modalities for a fast characterization of the inner ear in, e.g., gene knock out mice with unknown phenotype.

Techniques such as optical thin-sheet laser imaging microscopy (TSLIM) proved to provide high image resolution and contrast for inner ears of small animals like mouse or rat (Santi et al., 2009) and also human (Johnson et al., 2014) but require besides decalcification additional clearing techniques that may impact tissue morphometry more than our approach. Our assessment of tissue shrinkage caused by our chelate based decalcification and dehydration to 70% ethanol in order to get rid of air bubbles is surprisingly low. This may be attributed to the fact that we use a gentle way of decalcification with EDTA at 37°C in a neutral buffered solution and that the extended dual fixation with formaldehyde/glutaraldehyde and OsO₄ stabilized also delicate structures enough to resist medium ethanol dehydration. Developed initially for TEM we are able to rate the excellent cellular preservation from many previous studies on human temporal bones processed for scanning and transmission EM (Glueckert et al., 2005b; Rask-Andersen et al., 2012).

For very big human specimens, reduced voxel size for standard scans and enhanced processing time for decalcification did not meet our criteria for a higher throughput evaluation, so we decided for I₂KI contrast enhancement. The purpose for surgical planning of vestibular prototype electrode insertions required image data that include both middle ear and mastoid bone, as well as soft tissues such as blood vessels, dura and muscles. This was achieved with a voxel size of 25 μm and a field of view of roughly 50 mm. With this image resolution, even the smallest skeletal muscle in the human body – the stapedial muscle – could clearly be traced along its whole length (not shown) and the membranous labyrinth in the cochlea and vestibular system could be traced partially. Contrast is much lower than in decalcified and osmium-stained tissue (Figure 8A) so 3D median filtering and repeated scrolling through the image stack for structure recognition is indispensable. This makes it much more difficult to safely segment the membranous labyrinth and is also applicable only for bigger structures of the endolymphatic compartment. On the other hand, this minimal invasive technique without any manipulation at the round or oval window and avoidance of dehydrating agents may be suitable for electrode insertion studies that could be combined with histological techniques to assess mechanical trauma.

Manual segmentation of all the structures done in this study was very laborious but all methods so far tested to speeding up this process in the end failed to provide a precise tracing of delicate membranous structures. Model-based segmentation established on average image representation or multi-atlas segmentation utilizing segmented training sets may represent advanced ways to overcome many problems of manual fine structure segmentation. The latter method takes advantage of datasets of “atlases” (training images that have been previously labeled manually by an expert) (Iglesias and Sabuncu, 2015). An intensity template is registered non-rigidly to a target dataset and the resulting transformation is used to propagate the anatomical structure label of the template into the space of the target dataset (Lotjonen et al., 2010). For these methods data need to be known before and some information about anatomical variation is necessary (Fritscher et al., 2014). With our data on anatomical variation of soft tissue structures in the human inner ear we are now able to take the next step and to build up such advanced segmentation methods on basis of statistical shape models already established (unpublished data), aiming to reduce bias from manual segmentation. Combing shape and appearance model-based segmentation from microCT data with clinical CTs and high-field MRI may allow subpixel accuracy segmentation of certain structures (Demarcy et al., 2017). Information on the position and mean shape of bony and membranous labyrinth and sensory epithelia may provide relevant information for future surgery planning with augmented clinical 3D datasets.

Our preferential preparation technique of aldehyde-fixation and OsO₄-post-fixation offers another crucial advantage. Our specimens can further be processed for SEM, TEM, or 3D-EM methods such as serial block face SEM or focused ion beam SEM without restrictions in a correlative workflow. During the last years, microCT was used increasingly as a tool for scouting samples or for providing morphological reference for later investigation with light and/or electron microscopy (Handschuh et al., 2013; Sengle et al., 2013; Karreman et al., 2016a,b, 2017; Morales et al., 2016). For inner ear samples, microCT images could be extremely useful to detect anomalies or pathologies as we have shown previously in combinations with SRmicroCT and histology (Schmutzhard et al., 2009a,b; Glueckert et al., 2010). Subsequently, the region of interest can be precisely targeted which significantly speeds up any ultrastructural investigation. Furthermore, already established image registration protocols (Handschuh et al., 2013) would allow to put electron microscopy data at the nanometer scale in spatial 3D context of the whole inner ear geometry as provided by microCT. In future studies we plan to utilize electron microscopic analysis with exact tonotopical localization in the cochlea along with microanatomical and ultrastructural techniques to get a more comprehensive view of the fascinating and highly complex hearing and balance organ across a representative number of specimens.