Neuronavigation accuracy of the VOSTARS wearable AR platform vs traditional infrared navigation in a comparative study

Abstract

Neuronavigation is integral to modern neurosurgery. Clinical studies demonstrate its effectiveness. The primary tracking modalities in neurosurgical navigation are optical tracking systems (OTS) and electromagnetic tracking systems (EMTS). OTS remains the gold standard due to its accuracy and reliability. However, inherent inaccuracies due to brain deformation, image resolution, tool calibration, and registration errors can impact overall accuracy significantly, which differs from the system-declared accuracy. Augmented reality (AR) technologies solve traditional navigation challenges by integrating virtual information with the patient’s anatomy, enhancing the surgeon’s focus and cognitive load management. Head-mounted displays (HMDs) offer ergonomic benefits, although most AR-based neuronavigation studies have been limited to proof-of-concept trials. This study aims to evaluate VOSTARS, a novel hybrid video and optical see-through HMD designed for precision surgery, specifically in neurosurgical oncology for targeting supratentorial tumors. Previous in-vitro studies using patient-specific phantoms have shown promising results, with high accuracy in real-to-virtual target visualization and craniotomy trajectory tracing. With this work, we further assessed VOSTARS’ targeting accuracy within a realistic neurosurgery clinical workflow and compared its performance to the commercial StealthStation system on a patient-specific phantom. Our results demonstrate that users achieved the same median accuracy, 2 mm (IQR: 1 mm), over 60 measurements with both VOSTARS and the StealthStation with no statistically significant difference between the systems, confirming the non-inferiority of the VOSTARS platform compared to a commercial optical tracking-based surgical navigator.

Introduction

Neuronavigation offers several advantages to surgeons, and has been evaluated as the standard of care in clinical practice for precise planning of surgeries involving incisions and craniotomies, as well as for the identification of small subcortical lesions¹. Literature highlights its integration into routine procedures due to its effectiveness. Moreover, it allows the incorporation of functional magnetic resonance imaging (fMRI) and tractography data as overlays during surgeries^2,3. Clinical demonstrations of the neuronavigation benefit include the study from Wirtz et al.⁴ focusing on glioblastoma surgery. The study compared the impact of neuronavigation on time consumption, the extent of tumor removal, and survival. It showed that absolute and relative residual tumor volumes are significantly lower with neuronavigation, and patients who were operated on with neuronavigation also had longer survival. Additionally, neuronavigation increased the extent of tumor removal in glioblastoma resection without prolonging operating time.

In general, the neuronavigator functionalities show the surgeon the current pose of the surgical instruments in relation to the patient’s pre-operative volumetric images and/or surgical planning.

Two primary modalities are used in navigators to determine the pose of the surgical instruments: optical tracking systems (OTS) and electromagnetic tracking systems (EMTs). Both methods are widely employed in neurosurgical navigation, demonstrating their effectiveness and popularity. EMTs offer the advantage of not requiring a direct line of sight between the transmitter (electromagnetic field generator) and the receiver (sensor coil within the surgical tool). However, EMTs have limitations, such as the need to place the electromagnetic field generator close to the surgical site, a restricted field of view, and a lower tracking accuracy. The literature generally confirms that the OTS is still the gold standard for surgical stereotactic navigation, with a long track record of successful use in various types of surgery.

The visualization of the current pose of the surgical instruments relative to the patient’s preoperative volumetric images and/or to surgical planning can be improved via Augmented Reality (AR) technologies. AR visualization addresses the main issue of traditional navigation systems integrating virtual information directly with the patient’s anatomy, thus avoiding the surgeon’s need to frequently shift his/her focus between the surgical field and the navigation screen. For example, AR can show the position of an instrument with respect to a target structure as well as it can overlay a planned cutting line on the patient’s bone structure, guiding osteotomies more effectively without the need to lose the direct view of the patient while looking at the navigator display^5,6. This method reduces cognitive stress and improves information management during image-guided surgery.

In open surgery, head-mounted displays (HMDs) offer an ergonomic solution to provide AR by preserving the surgeon’s egocentric view of the surgical field. Thus, HMDs are deemed the most ergonomic and efficient tool for guiding surgeries that require direct manual manipulation of structures and tissues.

In recent years, there has been a significant surge in research dedicated to the development of AR-based neuronavigators.^7,8 with a focus on surgical rehearsal before the intervention, some access guidance for biopsies⁹ up to guidance for surgical resection of intracranial meningiomas¹⁰, providing insight into the disruptive potential of AR in neurosurgery. However, most studies on HMDs are ’proofs of concept’ trials based on using a Microsoft HoloLens¹¹, a self-contained Optical See Through (OST) headset, outside its indication and despite the technological and human-factor limits that prevent achieving high accuracy levels: the perceptual conflicts between real world and the VR image view^12,13, the small field of view (FoV), the sub-optimal ergonomics, and calibration issues to attain an accurate VR-to-real alignment.^5,14–16.

Despite some innovative devices arousing on the market for spinal surgery¹⁷, to the best of the authors’ knowledge, no head-mounted display specifically designed for brain surgery and compliant with medical device regulations has already been validated in a relevant environment for guiding high-precision tasks. This work focuses on the VOSTARS HMD, a surgical navigation platform developed within the Horizon 2020 European project framework^5,18. It has already demonstrated promising results in guiding maxillofacial osteotomies⁵. Furthermore, we validated its navigation performance in a neurosurgery set-up through a recent in-vitro study using a patient-specific phantom⁶ yielding impressive results, with a mean real-to-virtual 3D target visualization error (TVE3D) of just 1.3 mm and a standard deviation of 0.6 mm. Additionally, user studies showed that subjects guided by VOSTARS could trace a remarkable 97% of a planned craniotomy trajectory within a 1.5 mm error margin, with an outstanding 92% achieving a 1 mm margin. These results were obtained using a skin-fixed Dynamic Reference Frame (DRF) for real-time registration. The optical tracking algorithm that provides the pose of the patient-specific DRF relative to the HMD reference system relies on the stereo localization of three dyed spherical markers embedded into the DRF¹⁹.

In this study, we aim to thoroughly evaluate the targeting accuracy achievable in a realistic clinical workflow with the VOSTARS system when combined with the DRF and to compare its performance with results obtained using a state-of-the-art commercial navigation system, specifically the StealthStation system (Medtronic Inc., Louisville, CO, USA). Tests were conducted on a patient-specific phantom to ensure realistic and reliable results.

Methods

The StealthStation S7 System (Medtronic Inc., Louisville, CO, USA) is a widely used surgical navigation system offering real-time guidance by integrating radiological images with dynamic tracking of surgical tools. Utilizing either optical OTS or electromagnetic (EMTS) technology, it provides surgeons with the capability to upload patient-specific CT or MR images obtained preoperatively, or fluoroscopic images taken intraoperatively, which are then displayed from various angles including axial, sagittal, coronal, and oblique views. Surgeons can preoperatively plan and save multiple surgical trajectories and create detailed 3D anatomical models for enhanced visualization. During surgery, the system continuously updates the instrument positions on these images by tracking specialized surgical tools within the patient’s anatomy, ensuring precise and accurate surgical guidance.

In our work we decided to use the OTS tracking method, aiming to test our system against the gold standard²⁰.

The VOSTARS AR HMD and surgical navigation platform version 1.1 was created by mounting dedicated hardware on a customized version of a commercial OST visor (ARS.30 by Trivisio SAS, Chantilly,France ). The ARS.30 visor offers dual SXGA OLED panels with 1280x1024 resolution, a ${30}^{\circ }$ diagonal field of view, and a 3 cm eye relief. The OST display has an average angular resolution of approximately 1.11 arcmin/pixel, which is comparable to human visual acuity. The visor’s collimation optics were purposely redesigned to have a focal length of about 40 cm. Additionally, the two optical engines of the visor are slightly toed-in ($3.4^{\circ }$C), meaning that the optical axes of the two displays converge at approximately the focal length of the collimation optics. These features are crucial for reducing issues like vergence-accommodation conflict and focus rivalry¹² when the headset is used in the peripersonal space, as for the surgical field. VOSTARS can provide optical and video see-through augmentations using two liquid-crystal (LC) optical shutters (FOS model by LC-Tec Displays AB, Borlange, Sweden) placed on top of the semi-transparent optical combiners of the visor. Users can switch between a regular optical see-through (OST) view ( LC shutters open) and a video see-through (VST) camera-assisted view (LC shutters closed).

A pair of world-facing RGB cameras (two USB 3.0 LI-OV4689 cameras by Leopard Imaging (Leopard Imaging Inc, Fremont, California, US), both equipped with 1/3” OmniVision CMOS 4M pixels sensor (pixel size: 2um) and an M12 lens with 6 mm focal length) are mounted on the modified case of the ARS30. These cameras are used for inside-out tracking and to provide the VST view. The cameras offer a horizontal field of view of approximately ${46}^{\circ }$¹⁸, corresponding to an average angular resolution of about 2.2 arcmin/pixel. The stereo camera pair is mounted on the top of the visor with an anthropometric interaxial distance of about 63 mm to minimize the effect of camera-to-eye parallax. This setup achieves a quasi-orthostereoscopic perception of the scene under VST view.

For neurosurgical applications, we designed and preliminary tested a custom-made DRF to be used for registration during procedures⁶. The DRF mounts a frame of three colored fiducial markers embedded within the patient-specific area of the template. The DRF is 3D printed with the biocompatible and sterilizable material MED610 from a polyjet 3D printer (ObjetM30, Stratasys INC, MN, USA), ensuring patient safety. Pre-operative MRI scans guide the creation of this template, ensuring a customized fit for each patient’s face. The three colored fiducial markers are spherical (each 12 mm in diameter) and strategically placed to serve as tracking markers during the registration process, a design rule holds in placing the centroids of the spheres as close as possible to the intervention target without hindering the surgical gesture to maximize accuracy. The template provides a seamless registration process, thanks to its precise fit and the distinct positioning of the markers on the face. The form and location of the template are strategically crafted to align with facial structural features that remain stable, experiencing no or minimal deformation due to a slender layer of soft tissue beneath them.

Patient-specific phantom

Experiments were performed in vitro on a patient-specific 3D-printed head mannequin. The 3D model of the mannequin was generated from a real patient MRI dataset (an axial spoiled gradient recalled acquisition in the steady-state (SPGR) sequence with a 0.5X0.5X0.6mm resolution), segmented with a semi-automatic neighborhood-based region growing pipeline²¹ to extract the head surface. After segmentation and mesh preparation, 10 holes (3 points in the frontal region, 4 in the temporal region, 1 in the orbital region, and 2 in the nasal region) 1 mm in diameter were designed on the phantom surface to be used as targets for accuracy evaluation. Two DRFs were designed for the phantom in the PTC Creo 8 (PTC Inc., Boston, MA, USA - https://www.ptc.com, last access 18-06-25) CAD environment and used during the trials for the VOSTARS test session. According to the results reported in⁶, the two DRFs were designed to guarantee that the coloured spheres were as much perpendicular as possible to the surgeon’s viewpoint along frontal and lateral view, to minimize visualization error (Fig. 1).

Fig. 1.

PTC Creo 8 (PTC Inc., Boston, MA, USA, https://www.ptc.com ) CAD view of the dynamic reference frames (DRFs) designed for the experimental study and their respective landmarks. (A) The frontal DRF is depicted on the 3D mesh of a patient extracted from the MRI images using the EndoCAS Segmentation Pipeline²¹. This DRF was used to target the landmarks in the frontal and nasal regions. (B) The lateral DRF is shown. This frame was used to target the landmarks in the temporal and orbital area. NB The red targets shown in both figures are the landmarks designed at the canthi of the eyes used as a sanity check for the proper placement of the DRF while using the VOSTARS navigator.

Both the mannequine and the DRFs A fused deposition modeling (FDM) 3D printer (Fortus F170, Stratasys INC, MN, USA) was used to print the virtual model into an acrylonitrile butadiene styrene, ABS, replica.

Subjects enrollement

Six subjects (1 neurosurgeon, 5 biomedical engineers) aged between 27 and 41 with normal or corrected-to-normal visual acuity (with ophthalmic glasses or contact lenses) were recruited. Table1 reports the participants’ demographics to perform navigation trials with the VOSTARS system and the StealthStation S7. All the recruited subjects were instructed about the two navigation modalities, the procedure, and the data collection, and only after a complete rehearsal, they signed an informed consent. The subjects enrolled hold a medium-high level of competence in surgical navigation systems development and testing.

Table 1.

Demographic information.

Info	Values
Gender (male; female; non-binary)	(1;5;0)
Age (min; max; STD)	(27;41;33.17;5.98)
Visual acuity (normal; corrected to normal)	(5;1)
Experience with AR navigation in-vitro (none; limited; familiar; experienced)	(0;1;3;2)
Experience with commercial OTS navigation in-vitro (none; limited; familiar; experienced)	(0;1;2;3)

Phantom experiment protocol

Subjects were randomly assigned to use the VOSTARS system or the StealthStation first and were instructed to target the 10 mannequin landmarks (target holes) using the two navigators. In both cases, landmarks targeting was performer with the pointer of the StealthStation. The user dipped the pointer tip into a liquid dye and marked to mannequin following the navigator guidance, With this choice, we avoided possible biases related to the weight and footprint of the targeting tool. The pointer features a spherical tip of 1 mm size, matching the target holes.

Below the testing protocol for both navigation methods is reported. Only steps iii and iv differ according to the different navigation (V=Vostars; S=SthealthStation ) methods and divergences have been detailed in the list:

• iLine the phantom scalp with adhesive tape to hide each landmark position.
• iiMount the phantom head on the skull clamp.
• SiiiMount the head tracker on the skull clamp and perform registration.
• SivCheck the accuracy of the registration and repeat step Siii until the registration is successfully performed.
• ViiiPut the DRF on the phantom.
• VivCheck the sanity of the registration (canthi of the eyes, Fig. 1) and adjust the position of the DRF until the sanity check is satisfying.
• iiiInstruct the user to dip the pointer tip into liquid dye, target each landmark as shown by the navigator, and make a colored mark at each landmark.
• ivRepeat Step iv for each subject.
• vUse a sharp instrument to uncover the actual position of each landmark. Figure 2 shows an example of landmark targeting in the nasal area.
• viThree experimenter, blind to the navigation method, uses a caliper to measure the distance of each mark from the actual landmark and the mean value is recorded. For marks completely inside the holes (not visible on the adhesive tape after Step v), an accuracy of 0.5 mm is recorded.

Steps iii and iv for the StealthStation navigation relied on the Medtronic reference frame attached to the skull base holding the phantom head, Fig. 3. The StealthStation registration procedure was followed and refined until an error below 1 mm was achieved. In VOSTARS navigation trials, the DRF was positioned on the phantom’s face and held in place by elastic bands, Fig. 4. The registration was obtained implicitly, as the DRF fits the phantom’s face, providing pose registration. A registration “sanity-check” procedure was performed^20,31 to verify the proper placement of the DRF and its maintenance during the trial. 4 virtual spheres were designed at the canthi of the eyes and continuously projected in AR throughout the testing period. This allowed the experimenter to estimate the accurate template placement from different viewpoints visually. The positioning of the template was adjusted until the AR spheres appeared perfectly aligned with the corresponding anatomical landmarks.

Fig. 2.

Example of landmarks targeting in the nasal region.

Fig. 3.

Experimental setup for the StealthStation navigator. The optical tracker and the two monitors of the Medtronic navigator are highlighted by the yellow arrows. Within the red circle is the antenna tracked by the two IR cameras of the optical tracker, while the blue circle highlights the IR pointer of the StealthStation.

Fig. 4.

Experimental setup for the VOSTARS navigator. The head mannequin is mounted on the skull clamp (orange arrow), whilst the frontal dynamic reference frame (DRF) is secured on the mannequin using elastic bands (red circle). The user performs the landmarks targeting through the augmented reality images projected onto the VOSTARS displays (yellow circle). NB The StealthStation pointer was used in the VOSTARS experiment as well to avoid any possible distortion introduced by different dimensional characteristics of the targeting tool.

Statistical analysis

SPSS 22.1 (SPSS statistics, IBM inc., New York, NY, USA, https://www.ibm.com/products/spss-statistics - last accessed 18-06-2025) software was used to perform statistical analysis of data. Data were tested for normality using the Shapiro-Wilk test and the results of the targeting accuracy estimation were summarized in terms of median and interquartile range (IQR). A Wilcoxon signed-rank test was used to assess whether there was a significant difference in the users’ targeting accuracy based on the navigator used. A p-value less than or equal to 0.05 was considered significant.

Results

Table 2 summarizes all the results reporting the median value, IQR, and maximum errors. Overall, users achieved 2 mm median accuracy for both the navigation modalities with the maximum error measured of 5 mm for VOSTARS and 5.3mm for the StealthStation, respectively. Table 3 shows the performance results of six subjects who tested the VOSTARS system and the StealthStation for targeting 10 different landmarks on the phantom. These landmarks include three points in the frontal region (FR), four in the temporal region (TR), one in the orbital region (OR), and two in the nasal region (NR) (Fig. 1). The results of the Wilcoxon signed-rank test indicate no statistically significant difference in the overall VOSTARS vs Stealth performance. Single users’ performance was also tested confirming no statistically significant difference within VOSTARS vs Stealth performance, thus demonstrating the non-inferiority of the VOSTARS surgical navigation platform with respect to the IR based Stealthstation commercial navigator. Further data analysis revealed that the errors related to the landmarks in the temporal region were higher for both navigation modalities. This increase in error could be attributed to the IR tracker for the StealthStation being positioned far from the specific region, as shown in Fig. 5, and possibly the lateral DRF with the RGB markers not being sufficiently close to the operative area. A Mann-Whitney analysis indicated that the difference in errors was significant for the StealthStation (p < 0.01), but not for VOSTARS (p = 0.06).

Table 2.

Summary of the results.

Modality	Median [mm]	IQR[25;75]	MAX error [mm]
VOSTARS	2.0	[1.0;2.7]	5.0
StealthStation	2.0	[1.0;3.0]	5.3

Table 3.

Comparison of errors (mm) between VOSTARS and StealthStation S7 OT across all the landmarks and the subjects.

Landmarks	VOSTARS								StealthStation S7 OT
	S1	S2	S3	S4	S5	S6	median	IQR	S1	S2	S3	S4	S5	S6	median	IQR
F1	2.7	3.5	3.7	2.1	2.3	2.3	2.5	2.3	2.2	2.1	3.2	1.0	2.8	4.9	2.5	2.1
F2	1.5	2.8	3.7	0.5	0.8	1.0	1.3	0.9	0.5	0.7	1.1	1.0	0.9	1.5	1.0	0.8
F3	0.8	1.7	1.0	0.5	0.5	1.1	0.9	0.6	0.5	1.0	1.4	2.2	2.8	3.0	1.8	1.1
N1	1.2	1.2	2.1	0.5	0.5	0.5	0.9	0.5	0.8	0.8	1.0	0.5	0.5	1.2	0.8	0.6
N2	0.5	0.5	0.5	1.0	2.1	1.9	0.8	0.5	0.5	0.8	0.8	0.5	0.5	0.5	0.5	0.5
O1	2.0	2.3	3.5	0.5	2.9	3.0	2.6	2.1	0.7	1.6	2.7	2.0	1.4	1.4	1.5	1.4
T1	1.0	3.0	3.7	1.2	1.2	4.6	2.1	1.2	2.7	2.7	4.0	2.9	3.2	3.4	3.1	2.8
T2	2.3	3.6	3.6	0.5	1.5	2.5	2.4	1.7	2.7	3.2	3.8	2.0	2.6	3.2	3.0	2.6
T3	2.0	2.0	5.0	2.0	2.0	3.0	2.0	2.0	5.3	3.9	3.8	2.0	2.0	3.0	3.4	2.3
T4	1.2	2.7	1.2	0.9	1.4	3.6	1.3	1.2	3.7	4.3	4.5	1.7	2.2	2.2	3.0	2.2

Fig. 5.

To avoid encumbrances that would have hindered landmarks targeting, especially in the temporal region, the StealthStation tracker was placed on the opposite side of the mannequin head. It is reasonable to assume that the higher error in the temporal area is to be blamed on this increased distance between the pointer and the tracker.

Discussion

This study presents a comparative evaluation of the in-vitro operational accuracy of the innovative VOSTARS HMD AR navigation platform versus conventional IR-based optical navigation in neurosurgery. In particular, we compared the accuracy of the VOSTARS AR HMD and the StealthStation S7 optical navigation system in targeting superficial landmarks. Notably, the performance of both navigation systems was assessed within a clinically realistic surgical workflow. Accuracy was evaluated by directly measuring, using a caliper, the distance between the user-targeted point and the actual target. While the obtained values cannot predict the accuracy of the VOSTARS system in a real surgical scenario, the study design ensured that all procedural errors equally affected both navigation systems. Therefore, the results provide a meaningful comparison, demonstrating that the VOSTARS surgical navigation platform is non-inferior to a commercial optical tracking-based surgical navigator. The study underscores the importance of a well-designed DRF, considering the positioning of the RGB marker plane relative to both the surgical target and the surgeon’s viewpoint. This aspect is just as crucial as the proper placement of the IR tracker on the head fixation unit in traditional navigation systems, as both significantly contribute to minimizing overall navigation errors.

The accuracy of standard navigation systems has been widely investigated in the literature, allowing us to compare our results with other studies already submitted to the scientific community. For instance, in 2013, Koivukangas et al.²² evaluated the technical accuracy of the StealthStation S7 in a clinical setting using a custom-designed head phantom. The performance of both the Optical Tracking System (OTS) and Electromagnetic Tracking System (EMTS) was assessed within a 120times120times100 mm volume, approximately replicating the dimensions of the human head. The accuracy was calculated by evaluating the error in the distance of each target point to a reference point, both acquired with a navigated tooltip. The protocol does not require performing a registration procedure and avoids measured errors being influenced by human performance related to interface ergonomics and intrinsic user accuracy. Results showed that the technical accuracies of OTS and EMTS over the pre-determined volume are nearly equal: 0.20 mm ± 0.10 mm and 0.30 mm ± 0.13 mm, respectively.

It is worth underlining that estimated accuracies do not necessarily reflect true intraoperative accuracy: the actual accuracy of neuronavigation can be affected by various sources of errors. Specifically, Wang et al. identified two groups of errors based on the neuronavigator working principle. Type I error includes errors caused by differences between the anatomical structures in the diagnostic images and the actual patient, caused by brain deformation and image resolution. Type II error involves errors caused by the determination of the transformation of the position of tracked surgical tools from the patient reference frame to the image reference frame, including tracking errors of surgical tools and fiducials, surgical tool calibration inaccuracy, and image-to-patient registration errors. In general, the main contributors to the global error are related to the registration algorithm and fiducials’ mechanical instability on the patient. Regarding the latter, historically, neuro-navigation depended on direct matches of bone-implanted fiducials, offering high precision but causing patient discomfort. Now, systems have transitioned to paired point registration with adhesive, non-implantable markers, and surface matching algorithms with anatomical fiducials. Although these modern methods are slightly less precise than those based on implanted cranial markers, they are effective for routine use in most neurosurgical procedures. The global contribution of all the error sources ranges between 1.8 mm and 5 mm.

In the literature on wearable augmented reality solutions, the most commonly referenced device is the Microsoft HoloLens (Microsoft Inc.), a commercially available headset designed for general-purpose applications. Despite its widespread use in research, it is neither specifically designed for surgical use nor to be compliant with medical device regulations. Moreover, its design is intended for operation beyond the peripersonal space, which inherently limits its ability to achieve optimal accuracy in surgical settings^12,13. Clinical trials with HoloLens are often dedicated to surgical rehearsal. For example, in neuro-oncology, Jain et al. demonstrated that AR guidance can enable the surgeon to understand the relationship of the pathology with the surrounding structures²³. While focusing on operational accuracy evaluation, a study from Incekara et al. reports an accuracy of 4 mm around the center of a targeted tumor, where the “real” center of the lesion was determined through an optic neuro-navigator²⁴. These results were confirmed by a more recent study by Qi et al. that reports an overall median deviation of 4.1 mm (IQR 3.0 mm–4.7 mm) displaying lesion boundaries on the patient’s scalp again with respect to optic neuro-navigator²⁵. In Table 4, we report a survey of literature study about accuracy evaluation of AR-based navigation systems both with phantoms and patient study. The experimental setups and error metrics reported in the different studies vary significantly, making direct comparisons with our results challenging. Nevertheless, it is worth underlining that no recent study with OST HMDs reports errors below 3mms. While our results provide valuable insights, some limitations should be acknowledged. First, the accuracy measurements between the targeted and reached points were conducted manually using a caliper: this may introduce the potential for human error. To mitigate this, we employed three independent observers and averaged their measurements. A further improvement could involve each observer repeating the measurements three times to minimize both inter-observer and infra-observer variability. Second, the study was conducted on a phantom model with a relatively small sample size of six subjects. While this provided initial insights into system performance, a larger sample size would enhance statistical power, leading to more robust conclusions. However, logistical constraints, including limited access to the operating room and the StealthStation navigator, restricted the number of trials and participants. Future studies should aim to include a larger cohort to strengthen the statistical significance of the findings. Finally, the study did not investigate the impact of different template designs or variations in StealthStation tracker positioning on measurement accuracy. These factors could potentially influence navigation performance. A more comprehensive study, currently undergoing, will explore these variables in greater detail, including the use of custom-designed templates tailored to different lesion positions within the brain. This follow-up research aims to provide a deeper understanding of their impact on surgical navigation accuracy.

Table 4.

Survey of literature studies on AR-based navigation systems accuracy evaluation.

Category	Study	AR technology	Method of accuracy evaluation	Reported accuracy
Phantom study	Maruyama et al.26	Epson Moverio (BT-200) coupled with optical tracker	TRE over four target points at the border of a tumor	The mean and standard deviation were 3.1 and 1.9 mm respectively
	Demerath et al.27	Magic Leap	Euclidean distance to the target point measured with control CT	Median of 3 mm (IQR 1.7 mm)
	McJunkin et al.28	Microsoft HoloLens®	TRE over 7 pre-specified landmarks	Phantom Study: the mean and standard deviation were 5.76 and 0.54 mm respectively
Patient study	Incekara et al.24	Microsoft HoloLens	TDE measured using a BrainLab neuronavigator as a gold standard	The overall median deviation between the two modalities was $\approx$ 4 mm with an interquartile range 0-0.8 mm
	Van Doormaal et al.29	Microsoft HoloLens	FRE calculated as the root-mean-square of the distance between skin fiducials	Phantom Study: Mean and standard deviation were 7.2 and 1.8 mm respectively. Patients: Mean and standard deviation were 4.4 and 2.5 mm respectively
	Li et al.30	Microsoft HoloLens	Postoperative CT scan used to measure the TDE in guiding the external ventricular drain	Mean and standard deviation of 4.34 and 1.63 mm
	Fick et al.7	Microsoft HoloLens	FRE over 6 registrations	8.5 mm
	Qi et al.25	Microsoft HoloLens	TDE measured using a BrainLab neuronavigator as a gold standard	The overall median deviation between the two modalities was 4.1 mm (IQR 3.0 mm–4.7 mm)

Conclusion

This study provides a comparative evaluation of the in-vitro targeting accuracy of the VOSTARS augmented reality (AR) HMD and the StealthStation S7 optical navigation system. Both platforms achieved a median targeting error of 2 mm, with no statistically significant difference, confirming the non-inferiority of the VOSTARS system relative to the clinical gold standard. These results highlight the potential of dedicated AR solutions for neurosurgical guidance, and emphasize the critical role of an optimized Dynamic Reference Frame (DRF) in achieving accurate registration. Limitations include the use of a phantom model and a limited sample size. Future work will focus on clinical validation and further refinement of registration strategies. The development and clinical validation of dedicated AR HMDs that meet medical device regulations remain an essential step toward integrating augmented reality navigation into routine neurosurgical practice.

Author contributions

S.C. and M.C. conceived the experiments, supervised the work, and wrote the manuscript; M.C., and N.C. designed the experiments, M.C. and N.M. analyzed the results and wrote the manuscript; N.C. and M.A. conducted the experiments and prepared the figures and tables; F.C., V.F and E.C. participated in conceiving the research, S.C and N.M administered the project. All authors reviewed the manuscript.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethical approval

This research was approved by the Ethics Committee “Comitato Etico di Area Vasta Nord Ovest (CEAVNO)” (CEAVNO) on 14/04/2023 (Ethics Committee Opinion Register Number: 24035 MONTEMURRO). The research adhered to the principles outlined in the Declaration of Helsinki. All participants were thoroughly briefed on the study’s expectations and had to consent to the data-sharing and privacy policy before their involvement.

Informed consent

All individuals enrolled in the study provided their informed consent for the participation in the study, the acquisition and elaboration of the data related to the executed task, the evaluation of the related results, as well as the use and publication of possibly identifying information/images in an online open-access publication.

Competing interests

The authors declare no competing interests.