Three-dimensional histological specimen preparation for accurate imaging and spatial reconstruction of the middle and inner ear

Image-guided surgery (IGS) systems¹ or medical robots are gaining in clinical importance due to the increasing requirements for accuracy and safety of surgical interventions. These approaches in computer-assisted surgery (CAS) are mostly based on a three-dimensional (3D) image representation of the patient-specific anatomy as a necessity for preoperative planning and intraoperative implementation (Fig. 1). Hence, there are also growing demands in terms of adequate medical imaging. Especially in domains where high-resolution soft tissue imaging is necessary, imaging is a limiting factor for the implementation of a highly accurate computer-assisted surgical approach.

The implantation of a cochlear implant (CI) system for the treatment of patients who suffer from severe hearing loss or deafness is just such a surgical intervention which imposes high demands on intraoperative accuracy. There are various reasons why CI surgery seems to be predestined for computer-assisted improvements regardless of whether the research is focussed on one or more of the steps involved in CI surgery, such as:First, CI treatment is characterized by a high degree of social relevance owing to the large number of affected people and the central importance of hearing for human communication and thus social interaction. Improvements in hearing substantially affect the quality of life [19] and directly depend on the operative outcome. In addition, the indication for cochlear implants is changing; whereas it used to be total deafness, increasingly patients with residual hearing are being considered on the strength of motivating results with combined electrical and acoustic stimulation [10, 30, 60, 74, 104, 109, 110]. Thus, new challenges have to be tackled. In particular, the requirements for accuracy grow in the context of an atraumatic opening of the cochlea and electrode insertion. Furthermore, the lateral skull base shows a high complexity of functionally important anatomical structures which impose high demands on surgical precision and safety. Therefore, mechatronic devices and intraoperative visualization support the prevention of iatrogenic damage of endangered structures and also help to securely identify anatomical landmarks.

However, while procedures (i) and (ii) can be achieved by conventional radiological imaging, (iii) and especially (iv) approach the limits of currently available modalities. For these, adequate imaging of soft tissue structures is necessary for preoperative planning. During both the drilling of the cochleostomy and the insertion of the electrode array, no essential membranous structures within the cochlea should be destroyed. In particular, the preservation of the basilar membrane is crucial to preserve the residual hearing (see Fig. 2). The outer and inner hair cells are located on this delicate structure. They are the central functional elements of natural hearing, transducing acoustic information into auditory nerve response. The basilar membrane is situated approximately in the middle of the inner ear (cochlea), dividing the fluid-filled lumen and separating the scala tympani. It is of major interest in cochlear implant surgery as it is the preferred location for the electrode array. Thus, using computer-aided surgery to improve the precision and safety of the intervention, the shape and location of the scala tympani—which is partly bound by membranes—must be accurately known during imaging.

Additionally, development of cochlear implant electrodes on the basis of simulation and computer-aided engineering (CAE) is currently still rather uncommon. There are few simulation-based approaches to evaluating the available CI electrodes in terms of intra-cochlear movement and positioning [18, 46, 47]. Therefore, besides the minimally invasive approach to the cochlea, another aim of our research is to establish computer-aided design (CAD) in combination with simulation in the field of cochlear implant electrode development. Using finite element analysis (FEA), forces between implant and intra-cochlear tissues, such as the bony wall or the basilar membrane, will become calculable. Therefore, more complex and detailed models of the cochlear anatomy become necessary, which are based on adequate imaging that includes all essential soft tissue structures.

The challenge is to achieve high-resolution visualization of both high-contrast structures such as bone and soft tissue with nearly equal attenuation coefficients in close proximity. Since clinically available multislice computed tomography (MSCT), digital volume computed tomography (DVT) and magnetic resonance imaging (MRI) do not offer the required high spatial resolution and/or soft tissue differentiation, several imaging methods have been investigated. These experimental approaches include flat-panel-based volume computed tomography (fpVCT) and micro-computed tomography (micro-CT or \(\upmu \text{ CT }\)) as improvements on conventional CT imaging, magnetic resonance microscopy (MRM) as an experimental approach based on MRI, different (laser) light-sheet-based imaging methods such as orthogonal plane fluorescence optical sectioning (OPFOS) and optical coherence tomography (OCT), as well as different techniques of serial-section imaging. In this context, the literature often presents the basilar membrane as a benchmark for an adequate imaging outcome because it combines functional importance with the challenge of a thin soft tissue structure.

An important step in this context was the introduction of micro-computed tomography which has proven to be a suitable tool particularly with regard to high-density and high-contrast structures, such as the ossicular chain [21, 54, 55, 77, 84, 95, 106]. For the visualization of small structures, however, the thickness of the surrounding bone has to be minimized by specimen preparation because it absorbs a lot of radiation necessary for high-resolution imaging of interior structures [21]. Another option would be partial or complete decalcification of the bony structures [75, 80] or the use of less calcified foetal specimens [93]. However, achieving adequate imaging of soft tissue structures within the middle and inner ear is still a challenge and will require additional effort [68, 92]. The simplest approach is the optimization of scanning parameters with a special focus on soft tissue structures. Reduction tube voltage, for example, causes improved greyscale differences between soft tissue structures of the middle ear and background air [92, 95]. Moreover, long integration time and very low resolution would appear to enable the slight visualization of the basilar membrane as recently shown by Braun et al. [9]. Another possibility is the use of contrast agents. The injection of osmium tetroxide (\(\text{ OsO }_{4})\), for instance, was used to improve the visibility of the membranous labyrinth of the inner ear [76, 105]. In addition to the toxic osmium tetroxide, the use of low-toxicity contrast agents is reported for the imaging of vertebrate embryos [69], various animal tissues [68] and recently also for the human middle ear [2]. Alternatively, Postnov et al. [75] replaced the perilymph (the fluid which normally fills the scala tympani and vestibuli) with air to increase the contrast between the background and the basilar membrane for improved visualization.

Another effective way to improve the quality of X-ray imaging is the use of coherent hard X-rays from a synchrotron radiation source [31, 33, 56, 80] instead of portable scanners with their own X-ray source. This allows imaging of thin membranes even down to the subcellular range. The obvious drawback is the highly restricted availability of synchrotron facilities [68].

In addition, MRI was investigated as it offers obvious advantages in tissue differentiation in comparison with micro-CT imaging. Using high-field strength MR systems (additional local coils) and long acquisition time, resolution could be improved down to the micrometre range. Because the micrometre range is involved, these techniques became known as MRM. Beginning with Henson et al. [34] in 1994, 3D magnetic resonance microscopy was reported as an experimental method for imaging intra-cochlear soft tissue structures [34, 49, 55, 85]. When it was applied, the visualization of the basilar membrane with a thickness of about \(100\,\upmu \text{ m }\) and even of Reissner’s membrane (approx. \(10\,\upmu \text{ m }\) thick) became possible [49].

Another approach to visualizing anatomical structures non-destructively in high resolution is based on virtual optical sections. To generate these images, a thin laser light sheet is used and the omnidirectional fluorescence of the chemically treated specimen is recorded orthogonally [17]. This requires transparency of the object, either by using the natural transparency (as with some larvae or embryos [40, 41]) or by performing a chemical process (termed clearing) to render the sample translucent. Through rotation or translation of the specimen, series of 2D images from different directions or depths can be obtained and used for 3D reconstruction [37‐39, 88, 107, 108]. Voie et al. [107, 108] were the first to publish this new microscopic imaging technique. They called their method orthogonal plane fluorescence optical sectioning microscopy (OPFOS) and used it for visualization of the cochlea. Subsequently, improved middle and inner ear models were provided by Santi et al. [88] and Buytaert et al. [16] using OPFOS. Up to now, a couple of related or improved techniques have been developed and published under various names, including: selective plane illumination microscopy (SPIM, [40]), multidirectional SPIM (mSPIM, [41]), high-resolution (hr-)OPFOS [14‐16], digital scanned laser light-sheet microscope (DSLM, [45]), ultramicroscopy [22] and thin-sheet laser imaging microscope (TSLIM, [44, 87]). All these different implementations are covered by the acronym LSFM, which stands for light-sheet-based fluorescence microscopy. Excellent reviews are provided by [15, 17, 41, 86].

In the context of FEA of the middle ear including soft tissue structures, histological section preparation is used [27‐29] to overcome the above-mentioned drawbacks of conventional imaging. Sun et al. [101] published a very detailed and systematic description. This method was also used in several subsequent studies [27‐29, 102]. For a detailed reconstruction of the human middle ear, they decalcified a freshly extracted human temporal bone and fixed it in celloidin. Perpendicular fiducial holes were drilled for image alignment, and finally, the probe was serially sectioned. Sequential slices with a distance of 200 \(\upmu \text{ m }\) were used for 3D reconstruction. Wang et al. [112] applied the same preparation procedure but aligned consecutive images by means of an intensity-based registration algorithm instead of using reference markers.

Another detailed procedure description was published by Sørensen et al. [96] and performed for the visible ear project. Again, the tissue block (covering the temporal bone) was provided with vertical drill holes serving as position markers and afterwards sectioned in increments of \(50\,\upmu \text{ m }\). Image registration was performed by a least-square fitting algorithm on the registration markers. A more detailed overview on selected studies is provided in Tables 4 and 5.

These previously mentioned studies suggest that it is possible to completely reconstruct the complex anatomical structure of the temporal bone, including thin soft structures. However, \(\upmu \text{ CT }\) imaging is still limited in terms of soft tissue visualization, MRM does not provide the requisite resolution, and optical imaging methods such as OPFOS require a complex chemical procedure which influences the dimensional distortion of the tissue. Finally, commercial micro-CT scanners are expensive, and MRM and LSFM require custom-made experimental set-ups whose realization depends on wide experience obtained over several years.

An improved serial-sectioning method, based on our long-standing experience of the microgrinding procedure of the rigid fixation in epoxy resin, seems to have advantages regarding soft tissue visualization. Nevertheless, it still is a complex procedure dealing with several challenges, of which the minimization of dimensional distortion is the most demanding, as mentioned by Ferguson et al. [26]. This dimensional stability constitutes the precondition for geometrically accurate preservation of the anatomical structures, as the necessary preparation processes are destructive. The most remarkable fact about the above-mentioned studies using histological sectioning is that they do not include an investigation of the quality and accuracy of the performed 3D reconstruction. Therefore, it remains questionable whether these methods are really appropriate for accurate and realistic imaging, especially in the context of surgical interventions.

The motivation behind our study is to overcome these deficiencies. Since none of the clinically available imaging techniques offers either the desired spatial resolution or detailed soft tissue differentiation, we pursue a model-based imaging approach [89]. This involves combining standard medical imaging (e.g. MSCT) with digital anatomical models provided by a highly accurate imaging method which visualizes necessary soft tissue structures. By merging these data (which means warping the anatomical model to match the patient-specific images), clinical imaging can be completed by means of additional relevant information from a database containing averaged anatomical data.

For this reason, the aim of the present study is to provide a highly accurate preparation and 3D reconstruction procedure which is necessary for the further establishment of an anatomical database of this kind. As will become apparent, the method allows the dimensionally stable sectioning of relevant anatomical structures of the middle and inner ear (including the ossicular chain, corresponding muscles, tendons and ligaments) as well as of the membranous inner ear (including the stria vascularis, basilar membrane and scala tympani). Additionally, the accuracy of the 3D reconstruction method was investigated for the first time using a geometrically well-known reference object. Furthermore, reconstructed image stacks of embedded temporal bone specimens were merged with the corresponding imaging data from high-resolution computed tomography (fpVCT) to show the accuracy of the histological microgrinding procedure.

The proposed method was successfully performed on the reference object as well as on the embedded temporal bones. First, more general results concerning the accuracy of the introduced reconstruction procedure are presented; afterwards the findings from the temporal bone specimens are reported in detail.

This paper presents a highly accurate procedure for imaging of the middle and inner ear based on a serial cross-sectional preparation. To our knowledge, this is the first time a detailed investigation into the accuracy of a three-dimensional reconstruction procedure of this nature has been included. Therefore, a well-known reference object with comparable dimensions to those of the middle and inner ear was used [78]. As could be shown, the method allows the dimensional stable 3D reconstruction from 2D histological sections. This facilitates the detailed visualization of small anatomical structures including the ossicular chain with its muscles, tendons and ligaments as well as the membranous intra-cochlear structures such as the stria vascularis and basilar membrane.

The development of an accurate histology-based imaging technique was necessary because standard medical imaging techniques such as MSCT or MRI do not yet offer the necessary high spatial resolution and tissue differentiation to visualize thin soft tissue structures. Therefore, a strategy called model-based imaging [89] is pursued in our further research on CAS in the field of cochlear implantation as well as simulation-based optimization of CI electrodes. This concept has potential for bridging a gap in customized medical imaging by adding missing information, provided by a more detailed database. This database is to be generated from the provided histological image stacks obtained by segmentation of all relevant bone and soft tissue structures in a representative number of human temporal bone specimens.

To overcome the above-mentioned drawbacks of clinically available imaging, several methods were investigated in experimental studies. In the context of high-resolution imaging, micro-CT scanners in particular have attracted increasing interest. In 1999, Vogel [106] presented a new approach for a three-dimensional modelling of the human inner ear using both X-ray and synchrotron radiation. Spatial resolutions as low as 30 and 11 \(\upmu \text{ m }\) were achieved for the X-ray imaging and synchrotron radiation, respectively, which allowed adequate visualization of fluid and bony parts of the inner ear but provided no evidence of the basilar membrane or other intra-cochlear soft tissue structures. In 2004, Lane et al. [54] presented another study on imaging microscopy using a custom-made micro-CT scanner. By generating an isotropic data set of a cut temporal bone specimen with a spatial resolution of 20 \(\upmu \text{ m }\), they were able to depict bone structures in detail. Thus, as already underlined in previous studies, micro-CT proves to be a suitable tool for high-density and high-contrast structures such as osseous spiral lamina [97] or the ossicular bones [16, 21, 57, 77, 84, 95]. High-resolution ossicular chain models, for example, are presented by Salih et al. [84] and freely available on the Internet.

For visualization of very small structures, however, the surrounding bone has to be minimized by specimen preparation because it absorbs a lot of radiation necessary for high-resolution imaging of interior structures [21]. Alternatively, the temporal bone specimen needs to be decalcified by the use of inorganic acid to decrease the difference in X-ray absorption between membranous and bony structures [75]. Thus, micro-CT scanning is not always non-destructive, as noted by Decraemer et al. [21]. Additionally, imaging of soft tissue is difficult [16, 92, 97, 106] and can only be realized by special approaches to enhancing the soft tissue contrast. Postnov et al. [75], for example, replaced the perilymph with air to increase the contrast of the basilar membrane. This baseline scan was digitally superimposed with a second scan after replacing the fluid in the cochlear scalae. Although even Reissner’s membrane became visible, as mentioned by the authors, the influence of this procedure on the dimensional stability of the affected membranous anatomy is as yet unknown. An apparently less problematic method seems to be the use of contrast agents such as the injection of osmium tetroxide (\(\text{ OsO }_{4})\) [76, 105]. While osmium tetroxide is toxic and difficult to handle simpler and low-toxicity stains are described [68, 69] and successfully used for the human middle ear [2]. As for another drawback of micro-CT imaging—the noise of its raw data (especially in the context of subsequent segmentation)—the use of anisotropic diffusion methods for image denoising would appear to provide a suitable solution [76]. Finally, highly detailed anatomical models can be provided by combining micro-CT imaging with other modalities, as has been done by Buytaert et al. [16] for realistic middle ear modelling.

Focussing on soft tissue visualization, MRI is an object of research. Clinical application provides evidence that this method provides much better soft tissue differentiation than CT-based imaging, even though considerable restrictions in spatial resolution have to be accepted. Using high-field strength MR scanners and long acquisition time, the aim was spatial resolution in the range of micro-CT imaging [34, 49, 55, 85, 103]. Hence, it is often called magnetic resonance microscopy (MRM). These approaches are encouraging and help to promote a better understanding of complex human middle and inner ear anatomy.

Up to now, however, MRM has not provided the necessary high resolution for 3D modelling of the complex anatomy of the middle and inner ear [88]. Therefore, other non-destructive imaging methods have been developed, based on optical cross-sections of a transparent specimen. Although these techniques differ in detail, all use laser light which excites fluorescent light in the narrow region of a geometrically very thin laser sheet. A camera is focused on this thin laser sheet, processing only fluorescent light from the focal zone. In this way, 2D section images can be obtained [107]. By rotation or translation of the specimen, a well-aligned stack of images can be produced for further 3D modelling. All these related techniques are referred to as light-sheet-based fluorescence microscopy (LSFM) [15, 17, 86]. When virtual optical sectioning was first implemented, it was termed orthogonal-plane fluorescence optical sectioning (OPFOS) and used for imaging of guinea pig cochlea of [107, 108]. Subsequently this method has, for example, been applied to build-up a mouse cochlea database (MCD) [88] as well as a gerbil middle and inner ear model [16]. The MCD is available online and provides 2D serial image stacks as well as interactive 3D reconstructions of the ear.

As Table 4 show, complex procedures are necessary for specimen preparation, which include dehydration, decalcification, chemical steps to make the tissue transparent and the addition of fluorescent dye. Here, too, the resulting distortion and shrinking of the tissue influence the accuracy of the three-dimensional spatial reconstruction. An initial indication of the amount of shrinkage is provided by combining an OPFOS and a micro-CT model of the same ossicles from a gerbil specimen. A shrinking factor of 8.4 % was found [16]. Another challenge of OPFOS is the handling of neural tissue which cannot be rendered fully transparent and therefore must sometimes be removed [37]. However, for embryos and young mice (not older than two weeks), special clearing procedures are available that render complete brains (up to 2 cm in size) transparent [22].

The main advantage of the above-mentioned methods over histological serial sectioning is the non-destructive generation of serial cross-sectional images. Therefore, errors caused by loss of or damage to the slices, uneven slice thickness or inaccurate alignment can be excluded [97]. However, the low availability of the required technology is problematic. While micro-CT scanners are becoming increasingly commercially available, custom-made experimental set-ups are necessary for MRM and OPFOS. Furthermore, very specific know-how is needed on the part of the research groups, which normally has to be acquired over several years.

In comparison, a systematic and easy-to-follow procedure is presented which does not require expensive equipment. Additionally, it overcomes the mentioned drawbacks by providing visualization for both bony and soft tissue. Moreover, histological section preparation is still considered the gold standard for the evaluation of new imaging techniques as performed, for example, to evaluate the OPFOS [39] and MRM [71], or to compare the quality of MSCT imaging [117] and micro-CT imaging [75, 97], with histological sections.

Especially in the field of cochlear implant electrode evaluation using temporal bone specimens, serial cross-sectional imaging is still an essential tool with regard to the preservation of intra-cochlear soft tissue structures [1, 12, 48, 59, 82, 100, 113] as well as for the development of new imaging techniques for quality control after cochlear implantation [3, 4, 8, 43, 50]. For this purpose, the presented procedure opens up new possibilities, since it enables spatially accurate image fusion of post-experimental imaging and histological section preparation. To date, however, only an approximate comparison has been possible because corresponding views have to be chosen and aligned manually.

The most serious disadvantages pointed out by the published studies on histological serial cross-sectional imaging are artefacts caused by tissue preparation and the slicing technique, as well as errors due to inaccurate image alignment or non-equidistant slice thickness [34, 54, 97]. In fact, serial-section preparation can generally be seen as a procedure including some or all of the following steps, each of which is prone to errors: specimen preparation (e.g. dehydration, decalcification, fixation, trimming, embedding), sectioning (e.g. slicing, milling or grinding), digitalization (e.g. scanning, photographing) and several steps involved in image processing (e.g. image registration, filtering, segmentation, 3D reconstruction). Particularly noteworthy in this context is the fact that none of the studies about serial cross-sectional preparation known to us investigated the accuracy of the reconstruction procedure. This study is, therefore, focused both on the development of a highly accurate preparation and reconstruction procedure and its quantification.

This includes reducing the number of chemical steps in specimen preparation, especially the omission of decalcification. The preservation of bony structures as a rigid skeletal frame in combination with rigid embedding in epoxy resin also improves the spatially accurate reconstruction of soft tissue structures. Most of them are in close proximity and connected to bony structures which define and fix their spatial arrangement, such as the basilar membrane which stretches from the osseous spiral lamina to the outer bony wall via the radial fibres of the spiral ligament. This is an important advantage of the chosen procedure compared with non-rigid embedding methods such as embedding in paraffin. Although sectioning by microtome after paraffin embedding is also available in our laboratory, the introduced procedure was preferred because of the above-mentioned advantages. In our experience, it is not possible to perform an accurate 3D reconstruction of paraffin slices due to their inevitable wrinkling. Besides the advantage of a reduced risk of tissue deformation, other hard tissues (e.g. dental enamel) and metallic components of, for example, implants can thus be included in sample preparation that applies embedding in epoxy resin and microgrinding. Therefore, this procedure is of special interest for the evaluation of cochlear implant electrode insertion with regard to the determination of soft tissue preservation and the spatial relation between tissue and implant.

For image registration, external markers were chosen since perpendicular drilling using a CNC device provides the highest possible accuracy. An intensity-based registration algorithm was tested which, however, revealed an increasing registration error throughout the image stack, observable via the visible—though unused—registration markers. One reason for this might be the use of each image’s preceding image for alignment instead of a common base image. Hence, for further improvement of image stack alignment, a custom-made algorithm was developed which automatically detects the location of the registration markers via a two-stage process [78]. This enabled errors caused by manual image alignment to be excluded.

The most crucial aspect in performing serial cross-sectional imaging is to ensure equal image distances. A special specimen holder was therefore developed which allows precise definition of the desired abrasion via a fine thread and an abrasion-resistant hard ceramic ring. As a precaution, each abrasion was additionally measured. The analysis of these data shows that, in spite of the mechanical limitation of abrasion, a non-negligible error occurs. The measurement results provided an opportunity to compensate for this error during the construction of the Dicom headers. This was facilitated by the flexible Dicom standard which does not require equal image distance. Instead, the specific location within the stack can be assigned to each image. Possible reasons for the variation of abrasion distance might be the clearance of the thread and the manual pressing of the specimen holder on the grinding disc. It thus seems possible to remove material in the micrometre range even below the level of the hard ceramic ring.

Therefore, the determination of the de facto abrasion would appear to be necessary (as long as a more accurate adjustment and the achievement of equidistant abrasion are not available) for several reasons. Firstly, the error of abrasion is not uniformly distributed so that it is not compensated for stochastically. Secondly, proper measurement of the reconstructed total length does not allow the conclusion that all cross-sectional images within the stack are correctly located. As Fig. 7b shows, the error of abrasion remains undetectable without measuring. Thirdly, continuous measurement can improve the achieved accuracy since the presented results are based on the correction of insufficient abrasion by means of an additional grinding step. Fourthly, the deviation from equidistant abrasion will become more crucial if the abrasion is further reduced to enable more regular voxels.

The results achieved with the reference object confirm the practicability and dependability of the developed procedure. Deviations of less then \(0.4^{\circ } \,\pm \, 0.3^{\circ }\), taking into account shearing and tilting, show that only negligible distortion occurs during reconstruction. Likewise, an error in the range of 2.0 % for the reconstruction of lengths seems to be acceptable and tolerable for the underlying intention. Especially with regard to the aim of constructing an average anatomical atlas, these deviations are below inter-individual variations and thus compensated for by the averaging process. Comparison with outcomes of other 3D reconstruction procedures was not possible because, thus far, accuracy has not been reported in the literature. Although the results achieved with the reference object cannot be completely transferred to the embedding and reconstruction of tissue samples, the additionally performed image fusion confirms the consistency of the procedure even in this more complex context. Throughout all specimens, the discrepancy of the groove position within the fpVCT and the microgrinding image stack was not higher than 0.2 mm. Finally, the nearly equal shape of bony boundaries within the sample in both imaging methods can be considered admissible evidence for the accuracy of the introduced procedure. Remaining minor deviations are negligible because, finally, the approach of model-based imaging uses average values for several specimens, as mentioned above. Additionally, the intensity-based image fusion algorithm of iPlan is only performed on a limited area in the centre of the data set. Thus, the highest deviations occur at the image borders and are recorded by the measurement procedure. In the centre of the data, where the relevant anatomical structures are located, the error is very likely to be smaller. However, the use of fpVCT imaging as reference imaging modality is not an optimum solution as the resolution is limited. In general, it is recommended that the resolution of the reference standard is at least one dimension smaller than the investigated imaging procedure. This is not the case if comparing histological cross-sectional microgrinding preparation with fpVCT. For further investigations, the use of micro-CT imaging should be taken into consideration.

With an approximate pixel size of 16 \(\upmu \text{ m }\) in each cross-sectional image, the achieved imaging quality is high enough for adequate and detailed visualization of the relevant anatomy. Voxel size in the x-y plane can be increased still further by improving the magnification of the microscope and the resolution of the digital camera. In contrast, the voxel size in the z-direction (100 \(\upmu \text{ m }\)) is currently proportionally much lower, although it is twice as good as in fpVCT (approx. 200 \(\upmu \text{ m }\) isotropic voxel size, see Fig. 15) or highly improved compared with clinical MSCT (approx. 600 \(\upmu \text{ m }\) in the z-direction). As a consequence of the microgrinding procedure, voxels are not isotropic. This is a disadvantage for multiplanar reconstruction (MPR), as image quality will be altered. This applies especially to the basilar membrane, which is the most relevant anatomical structure but whose size is of the same order of magnitude. In principle, the differing edge lengths of a voxel can be equalized by decreasing abrasion distance, even though isotropic values are barely achievable. Unfortunately, the microgrinding procedure is very time-consuming. In addition to the already considerable amount of time needed for the preparation of one specimen with 100 \(\upmu \text{ m }\) abrasion, the necessary increase of resolution in the z-direction is another strong argument for automating the procedure. This particularly relates to the mechanical steps, including adjustment of abrasion, microgrinding, measurement and digital documentation. Taking these facts into account, the procedure requires an automated system comparable to the one used in [83].

Despite the quantitative determination of reconstruction accuracy and the carefully selected preparatory steps, it cannot be ruled out that the procedure influences the geometric dimensions of tissues. As the focus of this study was on the accuracy of the 3D reconstruction of sequential cross-sectional histological imaging, the chosen procedure does not allow the determination of the shrinkage and distortion caused by the specimen preparation steps, because fpVCT images were acquired after embedding. Therefore, this study does not provide information about the extend to which histological sectioning data deviate from the unprepared fresh specimens. This needs to be addressed in a future study utilizing additional imaging before commencing the specimen preparation and embedding procedure. For example, the shrinking of neuronal structures, this such as the vestibulocochlear nerve is conspicuous, this probably being caused by the dehydration. However, this is a common problem in all imaging techniques that require sample preparation. As shown in Table 4, most of the published procedures require one or more of the critical chemical sample preparation steps such as dehydration or decalcification. Likewise, if performing micro-CT imaging using contrast agents or OPFOS and rendering the tissue transparent (mostly via invasive perfusion through the round or oval window), integrity and dimensional preservation of the tissue cannot be ensured completely. Due to processes like these, none of the mentioned high-resolution imaging techniques can be applied to living humans. This underlines once more the importance of model-based imaging as a powerful tool to compensate missing anatomical details in the clinically available modalities (Table 5).

The presented imaging method based on a highly accurate serial cross-sectional preparation technique provides the groundwork for a new generation of middle and inner ear models. These include all necessary bone and soft tissue structures, such as are necessary for FEA. Furthermore, in the context of cochlear implant treatment, this procedure creates new possibilities for the evaluation of intra-cochlear electrode placement. By means of such an accurate correlation of CT-based and histological imaging during temporal bone studies, the quality management of cochlear implant surgery using clinical imaging can be further improved and established. Finally, complex models for improved implant development based on computer-aided design (CAD) and computer-aided optimization (CAO) can be generated from the provided image series using common segmentation techniques.

However, our prior motivation is the use of these data for segmentation and, finally, for the development of an anatomical database (atlas) of the human middle and inner ear. For this purpose, additional specimens shall be prepared according to the described procedure to calculate a statistically confirmed anatomy atlas employing average values from a representative number of data sets. This is the basis for the designated implementation of model-based imaging to achieve adequate preoperative planning of computer- and robot-assisted approaches within cochlear implant surgery. Only if the problem of imaging for preoperative planning is solved satisfactorily can highly accurate assistance devices be established for the surgical procedure, as mentioned in the Introduction. In particular, the accurate opening of the cochlea (cochleostomy) and the insertion of the electrode array into the scala tympani require adequate determination of the location of the basilar membrane during planning and performing of the minimally invasive procedure. Of course, the implantation of a cochlear device is only one example among a wide range of surgical approaches which can benefit from the three-dimensional, highly accurate histological imaging method, presented here.

There are multiple modalities in temporal bone imaging which are widely present in clinical routine and used for diagnostics in anatomical malformation, tumours or inflammatory diseases as well as for preoperative planning for interventions at the lateral skull base. However, imaging—of both soft and bony tissues—with high spatial resolution for highly accurate surgical procedures such as cochlear implantation remains a challenge. As part of our research in the field of cochlear implant surgery using mechatronic assistance, we pursue a model-based approach for preoperative planning to fill this gap in clinically available patient-specific imaging. Therefore, a database (atlas) of the human middle and inner ear should be generated using high-resolution images that include all necessary bony and soft tissue structures, especially the location of the basilar membrane.

For this reason, we developed and presented a highly accurate serial cross-sectional preparation technique. Accuracy was determined for the first time using a well-known reference object which has successfully proven its usability. The systematic and easy-to-follow preparation technique provided allows image fusion of the reconstructed image stack with conventional (imaging) modalities. The application of the procedure to human temporal bone specimens and its image fusion with fpVCT data underlined the spatial accuracy achieved.

The presented procedure opens up new possibilities in the fields of model-based preoperative planning, image-guided and computer-aided surgery, and biomechanical FEA by providing highly detailed and accurate imaging of both bony and soft tissue structures.