gms | German Medical Science

Navigation systems have found widespread use in otolaryngology [1]. The accuracy of navigation in clinical setup using MSCT is reported to be between 1.3 to 2.8 mm [2], [3], [4], [5], [6]. These accuracy values are accepted to be sufficient for frontal skull base procedures such as sinus surgery. Navigation systems are commonly used and highly supportive for these procedures [7], [8], [9]. However for lateral skull base cases these accuracy values are assumed to be too inaccurate due to the higher complexity of the petrous bone anatomy. IGS could be much more useful in lateral skull base surgery if its accuracy could be improved.

The accuracy of a navigation system is affected by various factors like the technical accuracy of the system, defined by resolution of the optoelectronic cameras, he type and quality of referencing and registration methods and the operative handling [10], [11], [12], [13]. The detail resolution of the underlying imaging has been previously identified by several authors to be the most influencing variable [14], [15].

A new kind of CT scanner is based on f lat p anel detectors with high resolution. As truly isotropic 3D imaging is acquired, this technique is called V olume C omputed T omography (fpVCT). Compared to routinely used up-to-date MSCT scanner it offers increased detail resolution of high contrast structures like bony skull base [16], [17], [18]. We hypothesized that the surgical accuracy of navigation systems improves significantly when using fpVCT data and that the gain in accuracy makes navigation assisted procedures at the lateral skull base feasible.

A prototype flat panel scanner manufactured by GE Healthcare (Milwaukee, WI, USA) was used in this study. The scanner consists of an X-Ray tube and two flat-panel detectors mounted on a standard CT gantry [14], [17]. The scan field of view is 12.8 cm x 12.8 cm using one panel and 33.5 cm x 33.5 cm in two-panel-mode. Each detector consists of 1024x1024 detector elements on an area of 20.48 x 20.48 cm², resulting in a detector element size of 200 x 200 µm². A 360° rotation took 8 seconds while 1000 projections were acquired. Z-coverage was 4.21 cm per rotation. A tube voltage of 120 kV together with a current of 40 mA were used. The reconstruction matrix of 512x512 pixels was selected as required for importing data by the navigation system. The axial reformations were exported in standard DICOM3 format. The measured, isotropic spatial resolution using a 25 µm Tungsten wire object (10% MTF, high resolution scanning and reconstruction) is approximately 25 line pairs per centimeter (25 lp/cm ≈ 200 µm feature size) for this scanner in a similar scanning mode.

The 16 slice CT GE Lightspeed 16 Pro build by GE Healthcare was used for comparison imaging. The high resolution scanning protocol includes 120 kV, 80 mA, 0.625 mm slice thickness and a pitch of 0.5 mm. The reconstruction was performed with a reconstruction field of view of 9.6 cm. A z-voxel dimension of 0.3 mm and the “bone plus” kernel in “standard” modus (180 weighted interpolation), resulting in a voxel size of 187 µm by 187 µm by 300 µm. The measured spatial resolution in plane and z-direction resulting using a 25 µm Tungsten wire object (10% MTF, high resolution scanning and reconstruction) is approximately 14-15 lp/cm (≈ 350 µm feature size) for this scanner.

A VectorVision² navigation system provided by BrainLAB (Feldkirchen, Germany) was used (passive optoelectronic locating technoloy). The iPlan software (version 2.0, BrainLAB) allowed manual identification of registration and control markers (part I and II of this study) as well as segmentation of functionally important structures and planning of a trajectory from the postauricular region to the basal turn of the cochlea (part II). The referencing adapters were attached to the phantom/cadaver specimen by a fixed friction-locked connection. Accurately defined holes (acrylic glass phantom) and titanium screws (human temporal bone specimens) served as fiducials for registration purposes (point match registration).

The custom made phantom was made of acrylic glass. Inverted cones were drilled at predefined positions by a CNC jig boring machine. The deepest point of multiple cones was marked manually in the MSCT/fpVCT datasets as fiducial points. During the registration process five points were localized with the pre-calibrated navigation pointer one after another. Seven other points were used for calculation of the Target Registration Error (TRE) [12], [13]. Each fiducial point for TRE measurement was localized by the pointer. The system calculated the distance between the navigated point and the marked position of the fiducial in CT-Scans. Screenshots in high magnification were taken for documentation. This procedure was repeated five times for each CT modality. The average TRE was calculated and compared for both imaging modalities by means of two sided t-test on depended samples (statistical analysis software: SPSS version 13).

Four human temporal bone specimens were prepared, each with five mini osteosynthesis screws (1.5 x 6 mm, Martin Inc., Tuttlingen, Germany) implanted as fiducial markers for registration (see Figure 1 [Fig. 1]). Functionally important structures were segmented, namely: the facial nerve and chorda tympani. A trajectory was planned from the retroauricular region going through the facial recess and tangentially entering the basal turn of the cochlea. As one specimen showed an exceptionally narrow facial recess the trajectory was planned to intentionally sacrifice the chorda tympani in order to preserve a sufficient margin of safety to the facial nerve.

During surgery, the specimen was fixed in a rigid fixture. A referencing adapter was attached to a medical drill (Aesculap, Tuttlingen, Germany) and it was calibrated as a navigated instrument. The RMSE value (Root Mean Square of Error) which is an indicator for the Fiducial Registration Error (FRE) was noted.

Initially, a 2.3 mm diamond drill bit was used. On the navigation screen the so-called autopilot view was displayed, showing a virtual corridor for the drilling procedure (see Figure 2 [Fig. 2]). Any deviation of the navigated drill from the optimum trajectory is displayed by arrows and numerical indicators. Furthermore the distance of the tip of the drill to the target point was visualized. The drill was advanced toward the cochlea while the navigation system was used to display the continuously updated position of the drill align the planed route to the cochlea. Every 10 seconds, drilling was stopped to prevent thermal damage to the nearby structures. Once the facial recess was passed and the drill entered the middle ear, the bit was changed to a 0.9 mm diamond bur for the cochleostomy. After recalibration, the procedure continued in the same manner until the target point was reached (see Figure 3 [Fig. 3]). The time of the preoperative setup as well as the time of drilling was noted.

Post procedure, a conventional MSCT imaging was acquired for visualization of the postoperative condition of the specimen (see Figure 4 [Fig. 4]). Secondly, a mastoidectomy and exposition of the facial recess were performed in typical manner.

The average TRE using the MSCT dataset was 0.82 mm (SD: 0.35 mm) with an error range from 0.3–1.7 mm. The average TRE using the fpVCT dataset was 0.46 mm (SD: 0.22 mm) with an error range from 0.1–1 mm. The accuracy of the navigation system measured on the phantom was significantly higher when using fpVCT instead of MSCT (p<0.01).

The calculated error of the registration consistency (RMSE) was 0.1 mm in 3 cases and 0.2 mm in 1 case. The time for preoperative planning was 35 to 85 minutes with a need for considerable training. The surgical procedure itself took about 10 to 15 minutes with intermitent pauses in order to minimize thermal stress on nervous structures in the vicinity.

Pre- and postoperative imaging as well as the virtual trajectory were superimposed, allowing for comparison between the planned surgery and the result. In the postoperative images the intactness of relevant anatomical structures could be controlled.

The surgical exposition of the mastoid and the facial recess (angle between facial nerve and chorda tympani) revealed in all cases a straight bony canal with smooth walls appropriate for the insertion of a delicate cochlear implant electrode. According to the preoperative plan the chorda tympani had been sacrificed in one case due to a very small angle between the chorda and facial nerve. Other anatomical structures such as facial nerve, carotid artery, sigmoid sinus, vestibular organ, middle ear ossicles or dura were intact in all cases.

The accuracy of a navigation system and its long term stability is the most important factor for indicating the intraoperative use of navigation systems for various procedures. Using state-of-the-art MSCT imaging and currently available optoelectronical navigation systems the surgical accuracy is about 1.3 to 2.8 mm [2], [3], [4], [5], [6]. These conventional navigation setups can be used therefore mostly in the region of the anterior skull base and sinus surgery in addition to current intraoperative techniques utilizing endoscopy or microscopy [19], [20], [21].

The lateral skull base and especially the petrous bone is characterized by an enormously dense anatomy. Conventional navigation systems do not provide sufficient accuracy for use in this anatomic area. A cochlear implantation (opening of the inner ear and insertion of a multichannel stimulation electrode for treatment of deafness or severe hearing deficit) is a frequent surgical intervention at the lateral skull base and demands high accuracy. Schipper et al examined the impact of the navigation systems on cochlear implant procedures and defined the necessary TRE of the navigation system should not exceed 0.5 mm for performing the cochleostomy [20], [21]. Cadaver studies revealed deterioration of the basilar membrane and dislocation of the cochlear implant electrode into the vestibular scale of the cochlea in about 20% to 50% of the cases in conventional cochlear implant surgery.

A re-design of the surgical procedure in terms of a) preservation of the inner structures of the hearing organ, b) optimal localization of the cochleostomy and c) atraumatic insertion of the stimulation electrode would result in a direct benefit to the patient. These objectives are of special interest in cases of Electro-Acoustic Stimulation (EAS) (use of a cochlear implant and a conventional hearing aid on the same ear). EAS aims on optimum use of bionic and residual natural hearing. It is a relatively new concept which is now supported by a handful of very promising clinical studies.

The accuracy of the navigation systems depends on a variety of factors including limitations of accuracy depending of the manufacturing of the device (technical accuracy) as well as errors related to the handling of the IGS system like repositioning the reference adapters attached to the patient's head or instrument during the procedure by mistake. The best intraoperative accuracy can still not surpass the 0.35 mm mark since this is the best provided technical accuracy of the current localization camera systems (Polaris, NDI, Toronto, Canada) which is the core technology of most optoelectronic medical navigation systems. As mentioned above detail resolution of the underlying 3D imaging is one major factor for navigation accuracy. We have shown that the quality of a navigated procedure can be improved significantly if fpVCT is used [14]. The fpVCT technique provides a voxel size of 200 x 200 x 200 µm³ and provides a higher detail resolution compared to 350 x 350 x 350 µm³ resolution of the MSCT [14], [16], [17], [22].

During the registration procedure, the corresponding position of the fiducial markers in the real world and the volume dataset need to be matched. Fiducial localization errors (FLE) inevitably occur both at image acquisition (FLE_I = image markers) and at every calculation cycle of the navigation core unit (FLE_P = physical space markers) [12], [14]. They can not be measured but can be approximated by means of localization systems with much higher accuracy such as electromechanic or laser-based localization devices in industrial use.

The registration process itself is the alignment of the two coordinate systems of imaging and navigated space. It is associated with the Fiducial Registration Error (FRE). Just like FLE_I and FLE_P, FRE cannot be measured. But FRE can be estimated by means of RMSE calculation. Usual RMSE values in clinical use range from 0.3 to 1.2 mm.

One element of error that can be calculated is the positioning error within the intraoperative site at the region of surgical interest or the target registration error. The resulting TRE values are often 2-4 times higher than the corresponding RMSE. The better resolution of the fpVCT imaging decreases the FLEI and therefore effects FRE and TRE.

In part I of the presented paper the TRE was measured. TRE summarizes all potential navigation errors and describes the offset of a definite point in the real world compared to its calculated position in the image dataset. By simply changing the underlying CT images from MSCT to fpVCT in a commercially available navigation system, the TRE is decreased from 0.82 mm (SD: 0.35 mm) to 0.46 mm (SD: 0.22 mm) in a phantom.

The results of the phantom test strengthen our hypothesis that the navigational accuracy at the lateral skull base could benefit from the use of fpVCT data and thereby become a safe standard of care procedure. The RMSE values of 0.1 and 0.2 mm in part II of this study indicate the significant accuracy improvement which we quantified in part I. Accordingly we were able to demonstrate that using fpVCT imaging on a navigated procedure makes a minimally invasive mastoidectomy with cochleostomy feasible. Postoperative CT imaging as well as conventional dissection of the temporal bones (mastoidectomy and exposition of the facial recess) showed that the drilled channel to the cochlea did not damage any anatomical structures and that the cochleostomy was placed loco typico. Only in one case we were forced to sacrifice the chorda tympani due to a very narrow facial recess. This was evident during the preoperative planning and the planned trajectory was intentionally set to pass through the chorda tympani in order to keep a margin of safety of 1 mm to the facial nerve, thereby ensuring its protection.

A similar approach was done by Labadie et al. [23], [24] using a maxillary occlusion splint with attached fiducial markers as a referencing and registration object. A channel was drilled to the middle ear but a cochleostomy was not performed. By using MSCT data the TRE was found to be 0.76±0.23 mm. With this study, all cochleostomies was performed loco typico, thus the real inaccuracy and thus the impact of the navigation system could not be determined as there was no visible deviation when compared to the preoperative planned procedure. Thus it can be assumed that the TRE is less than 0.5 mm in all these cases according to Schipper’s postulation.

The use of high resolution fpVCT imaging on navigated procedures can decrease the Target Registration Error to less than 0.5 mm so that high-accuracy interventions at the lateral skull base become feasible using state-of-the-art navigation devices.

The concept of a percutaneous approach to the middle ear via facial recess and navigated cochleostomy for cochlear implant surgery was proven here to be practicable. This is preliminary work but its translation into clinical application seems to be possible. There are still different concerns about this minimally invasive procedure which need further investigation such as the issue of the control of intraoperative hemorrhage. Another problem which is the matter of current promising investigations is the development of adequate instruments for insertion of the electrode through the minimally invasive approach. The use of medical robotic assistance could possibly ensure an optimum electrode positioning with positive effects on the patient’s audiologic outcome.

The authors would like to thank BrainLAB AG, Feldkirchen for supporting the program by making available the planning software. This paper was presented at the 5^th annual meeting of the German Society for Computer and Roboter Assisted Surgery (CURAC).

None declared.