Novel microscope-based visual display and nasopharyngeal registration for auditory brainstem implantation: a feasibility study in an ex vivo model

To enable spatial localization within the 3D space of the patient’s anatomy, a titanium medical forceps (B. Braun AG, Melsungen, Germany), typically used for grasping brainstem electrodes, was navigated using electromagnetic tracking (EMT). A 6D magnetic sensor (0.8 mm x 9 mm, NDI - Northern Digital Inc., ON, Canada) was attached on one tranche of the forceps; the instrument position was pivot-calibrated [11] at the tip. Figure 1 illustrates the setup.

Preoperative radiological imaging needs to be registered to the intraoperative patient setting to allow intraoperative navigation. An automated nasopharyngeal point-based registration that utilizes titanium spherical fiducials combined with EMT systems serves this purpose. A titanium nasal stent based on the AlaxoStent (Alaxo GmbH, Germany) normally used for breathing enhancement in patients [12]—served to elaborate minimally invasive positioning and stabilization of four spherical fiducials and isocentrically mounted magnetic position sensors in the nasopharynx prior to preoperative imaging (Fig. 2). An algorithm was developed and verified to automatically localize fiducials in preoperative imagery and match these positions with the intraoperative positions provided by integrated magnetic sensors [13].

The visual display (introduced in [9, 10]) indicates the displacements between the current position of the tracked instrument tip and the designed target point using intuitive, minimalistic, and semi-transparent visual clues superimposed onto the video of the actual surgical site. The three axes of the instrument were encoded with respect to the observer’s viewpoint: the z-axis runs longitudinally along the instrument tip while the \(x-y\) axes form a plane perpendicular to it, Fig. 1. The z-direction represents moving in/out and is termed as “depth”; movements in the \(x-y\) plane left/right and up/down and are termed as “lateral.”

The lateral and depth vectorial displacements between instrument-tip and target position were measured and visually represented in a 2D view (e.g., a microscopic view) (Fig., Supp. 1).

The lateral displacement was visualized with a green blob 2D-translated from a red square, the target (Fig., Supp. 1a-d). When the lateral distance is zero, the two visual cues overlap (Fig., Supp. 1d).

The distance in the depth direction is visualized as an interactive circle with a diameter proportional to the depth distance with its origin on the target. In other words, the circle minimizes or maximizes its diameter if the instrument tip is moving toward or away from the target (Fig., Supp. 1e-h). When the depth distance is zero, the circle vanishes (Fig., Supp. 1h). If the instrument is beyond (or overreached) the target, the circle behaves similarly but blinks to signalize for this (Video, Supp. 3, 07:27).

The virtual cues in pixels are scaled to the surgeon’s preferences to provide an intuitive perception of physical dimensions, such as revealing and emphasizing small instrument motions.

When the instrument is out of the defined target range, only the visual cues for target and depth are displayed, and the lateral displacement becomes lateral orientation with a long arrow pointing toward the target (Fig., Supp. 2). Furthermore, the diameter of the depth circle is kept constant.

An example of the visual display is shown in Fig. 3.

Navigation is perturbated by random noise and measurement errors [14, 15]. Therefore, intraoperative awareness of the accuracy range (“reliability”) is beneficial and could help avoid tissue damage and optimize electrode placement by avoiding directions with higher uncertainty when using the surgical instrument. We modeled these errors as the distribution of TRE (target registration error) [16] to quantify the uncertainty of instrument-tip localization at the surgical target point. The model was implemented using a general approach for first-order approximation [17] and was validated across two other similar algorithms [16, 18]. The chi-square distribution with 95% confidence regions are shown as green and red “error” ellipses in the lateral directions around instrument-tip and target (Fig., Supp. 1a-d).

The experiment workflow depicted in Fig. 4 is described below.

The mean deviation in fiducial positions from its base locations was \(0.29\,\pm \,0.20\) for the left-side intervention and \(1.31\,\pm \,0.30\) for the right-side intervention. A more detailed accuracy inspection for individual fiducials was reported [25].

The mean ± standard deviation quantities were measured for the left and the right side intervention. The FRE (fiducial registration error) [16] for six repetitions during usability (dummy) and final positioning gave \(0.53\,\pm \,0.1\) mm and \(0.44\,\pm \,0.04\) mm. The TRE measured on screws was \(2.56 \pm 1.27\) mm and \(2.87\,\pm \,0.87\) mm. The numerically predicted TREs on screws using the process described in Sect. 2.5.4 obtained higher values compared to the measured TREs (see Table 1). T2 screw was loosely fixed during evaluation. T1 and T8 were lost.

Table 2 presents the evaluation performed for dummy electrode positioning. IGS and VG contained 4 collected trials each while IGS+VG only 2.

Table 3 gives the errors for final electrode positioning. Figure 6 visualizes the distance between actual and planned target positions. With one exception, in the right ear side intervention, the error was assessed for the position determined using the instrument after fixation of the implant (Video, Supp. 3, 08:16) as it was found that the electrode was displaced from the planned position (Fig. Supp. 5; Table, Supp. 6). Figure 7 shows a comparison between final placement errors against usability placement errors independently along each axis calculated in the CT coordinate system (Fig., Supp. 7).

An ABI involves complex neurosurgery to the critical areas of the brainstem and cumbersome anatomical localization of the CN. In the literature, the reported outcomes vary considerably between patients despite intraoperative assistance of EABRs [2]. Even slight electrode misplacements, among other hazards, contribute to poorer outcomes [6]. Enhancing spatial orientation might help reduce the risk of the latter and assist when anatomical landmarks are limited due to the surgical approach, surgeon’s experience [6] and, particularly, anatomical variations that arise from NF2 tumor removal [23].

In fact, in many other medical procedures, less invasiveness, accurate target localization, and reduced risk of complications were reported by physicians when using neuronavigation (or image-guided surgery) systems [26‐28]. We hypothesize that these systems might offer solutions to surgeons confronting challenges during the conventional ABI approach, including more predictable and consistent surgical and audiological outcomes. However, a review of the literature reveals no prior use of these systems during ABI implantation.

We therefore demonstrated a microscope-integrated visually assisted framework that combines an intuitive and helpful guidance mode as well as highly accurate and automatized image registration. In particular, we used the same surgical instruments and electrode pads as in the clinical scenario. Nonetheless, they are not optimal for navigation precision, so certain design improvements could lead to still further improved results. However, it would subsequently require reiterating product and surgical tool development in compliance with medico-legal regulations without adding discomfort to surgeons.

The errors in final (1.58/3.16 mm) and usability (\(2.32\pm 0.61\) mm) positioning appear to follow the same trend along single axes as shown in Fig. 7. It seems that both lateral and depth deviations meaningfully describe the error in their respective directions, as potentially both are susceptible to human errors given that the electrode is placed onto a curved anatomical surface.

There are no direct comparisons with our experimental results. In general, they correspond to the reported accuracy of intracranial electrode placement in stereoelectroencephalography where a mean target error of 2.89 mm (2.34 - 3.44 95% CI) was observed across several frameless stereotactic systems [27]. The intracranial and ABI electrodes are not directly comparable. The deep electrodes typically have a diameter \(< 1\) mm and follow a predefined trajectory through the brain.

In line with our previous phantom study [10], the proposed visual interface achieves a lower localization error than cross-sectional views despite condensed information presented. However, slightly larger completion time and trajectory were observed. The surgeons, indeed, stated that guidance directly in the microscope was their preferred interface for the task at hand. They appreciated the possibility to be able to focus on the intervention itself while being guided toward the target with simple, intuitive visual cues that require minimum distractions and do not pose high cognitive load.

Nasopharyngeal registration was central to establish reproducible and accurate spatial navigation in the vicinity of the CN. Our code enhances surgical workflow and reduces errors induced by human interactions. This is of special importance to maximize reproducibility due to the so-called silent loss of accuracy phenomena that occurs during the course of surgical procedures [29].

An early concept of this scheme was introduced for lateral skull base surgery and yielded submillimetric accuracy in a phantom setup [30]. We extended this scheme with a novel device [12]. Contrary to gold standard of invasive skull-implanted fiducial screws that need to be distributed around the head to reach optimal TRE, we aimed to bring fiducials in a minimally invasive way as closely as surgically possible to the target in order to achieve high registration and targeting accuracy at the surgical target.

We report a maximum (intrasurgical) dislocation of the spherical marker, \(<1.72\) mm, which is in agreement with a recent endonasal magnetic tracker (\(< 2\) mm) [28]. As reported in the clinical setting, this is better than a skin adhesive tracker (\(< 3\) mm) [28]. The mean±standard deviation TRE measured on the implanted screw targets (\(2.71 \pm 0.99\) mm) is higher compared to registration with screws (\(<1.0\) mm [29]); approximately equal to adhesive marker registration (\(2.49 \pm 0.86\) mm [31]); and lower than surface matching-based registration (\(5.35 \pm 1.64\) mm [31]) as observed in clinical setup. However, higher TREs are to be expected in the current setting as the TRE grows proportionally to the distance of the target to the barycenter of the fiducial configuration [16]. The target points like the surgical point of interest were closer to the fiducial points, and thus TRE, as confirmed by numerical simulations (approx. 1.76 mm), can be expected to be lower.

Freehand navigation is advantageous in human-operated interventions, especially in the head-neck part, where the operating areas are small and cluttered [34]. One source of errors in our results is in part due to EMT. It yields inferior accuracy to optical tracking and is prone to ferromagnetic distortions present in operating theaters that could drastically degrade the accuracy [33]. Albeit, by prudently positioning the emitter, low-submillimetric accuracy can be observed in a laboratory [34] and clinical environment [33].

This study has limitations. First, even though the implant has a plate geometry, only a single target point at the tool tip was covered during navigation in the visual display. We attempted to compensate for these by additional positional quality assurance checks on the implant surface after positioning. Second, since it is difficult to measure targets in the brain accurately, we rely on theoretical approximations of the TRE [16, 17]. There is a fair agreement between the measured and the predicted TRE within the error bounds (mean ± standard deviation) on the screws. The difference between the two TREs could be attributable to incorrectly estimated FLE. Our analysis shows that potentially the FLE is overestimated. The FLE can be calculated by [35]: \(<{\mathrm{FRE}}^2>=(1-2/N)<FLE^2>\), where N is the number of fiducials and \(<.>\) means “expected value of”. Our results are also limited with only two performed interventions and thus precluding detailed statistical analysis. The study does not compensate for the influence of cerebellar retraction and brainstem shift during surgery, which can exacerbate errors in registration and navigation. Further, the outer nose had to be cut laterally to the cartilaginous septum to allow placing the Rhinospider stent; conservation of the specimen changed the elasticity of the tissue so that the device could not be inserted into the nasopharyngeal space without risking damage. The embalmed specimen brain tissue can be very rigid, and surgical exposure of the CN without damage could be cumbersome.

From this point, we believe that certain improvements would optimize surgical workflow. Any suitable atlas could be fused on the brainstem imaging to mark the nuclei more accurately, as this is routinely done for deep brain electrodes [36]. Furthermore, drawing from a recent study [37], by optimizing pre-operatively positioning of the entire dimensions of the electrode pad in the fused CT-MR images and predicting the categories of auditory performance or side effects resulting from non-auditory sensations before ABI activation. CT-MR combination might indeed be utilized for this task [2] as CT allows high-quality resolution of bony landmarks [37] while MR could provide better visibility of neural structures in the brainstem. Finally, bearing in mind that the electrode orientation contributes to auditory performance or side effects [37], estimating and presenting an intraoperative pose of the electrode array in the visual display could improve the positioning accuracy. However, this may be technically challenging with the current design of the electrode pad and forceps surgical instrument.