The PubMed, Scopus, Web of Science Advance, and Embase (Ovid) databases were retrospectively queried to identify all relevant studies from inception to March 28, 2024, using a search string with the following keywords: “extended reality,” “augmented reality,” “virtual reality,” “mixed reality,” and “skull base surgery” (Table 1). Individuals of all ages were considered for inclusion. Studies that were either unrelated to XR or did not include patients undergoing skull base procedures were excluded. Animal studies, reviews, and nonoriginal research articles were also excluded from analysis.

The initial title and abstract screening was led by 2 authors (V.S. and E.N.). Any discrepancies were addressed through consultation with a third co-author (A.B.) and further discussion with all authors. The screening of studies adhered to Preferred Reporting Items for Systematic Reviews and Meta Analyses guidelines (Figure 1).

Three independent authors (V.S., E.N., and A.B.) extracted relevant data from the selected studies. The data collected included study design, participant demographics, and the number of participants with respective outcomes and complications. Discrepancies in data extraction were resolved through consensus, and any unresolved disagreements were addressed through consultation with a fourth reviewer (A.T.E.). The results are summarized and discussed on the basis of the salient features extracted from the studies in the AR, MR, and VR categories. We included ventriculostomy and external ventricular drain simulations under “skull base surgery” due to their anatomical proximity to the skull base and their relevance in training to manage complications associated with skull base interventions. However, due to the limited data and insufficient published research, we were unable to determine the statistical significance among the various XR technologies and performance outcomes in skull base neurosurgery.

A total of 357 articles underwent initial screening, after which 107 duplicate articles were identified and subsequently removed. Careful evaluation of titles and abstracts yielded 29 articles that were deemed suitable for full-text examination. A detailed assessment excluded 8 articles; thus, 21 studies met the inclusion criteria and were incorporated into our study (Tables 2–4) (15–35).

Of the 21 studies included, 13 focused on AR, and 4 each focused on VR and MR. The detailed characteristics of each study are presented in Tables 2–4 (15–35).

Blending AR with real-world intraoperative imaging and navigation systems allows visualization of critical neuroanatomical structures in real time and can aid in the resection of skull base pathologies (36). For example, in a report from 2018, Umebayashi et al. described employing AR-enhanced navigation maps to guide electric drills during procedures (37). Similarly, in 2020, Jean demonstrated the use of 3-dimensional (3D) VR renderings for preoperative rehearsal and intraoperative guidance in a minipterional craniotomy approach, highlighting AR's capacity to improve surgical accuracy and outcomes by displaying the anticipated surgical opening as an overlay over the patient in real time (38).

In 2023, Goto et al. evaluated a novel AR navigation system for endoscopic transsphenoidal surgery, and its accuracy was measured in 15 consecutive patients using a 5-point scale, on which a 4 represented “as useful as conventional neuronavigation,” and a 5 represented “more useful than conventional neuronavigation.” In their study, the AR navigation received a mean score of 4.7 (95% CI: 4.58–4.82), indicating that AR navigation was more useful than conventional navigation in most cases. However, depth perception of the lesion was more difficult with AR navigation in 2 cases (a skull base chondrosarcoma and craniopharyngioma) (18). In 2023, El Chemaly et al. assessed stereoscopic calibration for AR in microsurgery, in which the intrinsic (e.g., principal point, focal length, lens distortion) and extrinsic (location and orientation) parameters of the surgical stereo microscope were calculated using a calibration board that was 3D printed and electromagnetically tracked. The electromagnetic tracker coordinate system and the stereo microscope image space were transformed using these parameters such that any tracked 3D point could be projected onto the left and right visuals of the microscope video stream. Calibration was done once the system was set up, and an average calibration workflow time of 55.5 s and an average computational time of 10.5 s were reported, both being critical components of operational efficiency (16).

The microscope can be efficiently precalibrated at multiple focal lengths and the corresponding intrinsic parameters can be obtained. Zooming or changing the focal length of the microscope necessitates recalibration; thus, algorithms that provide automatic camera parameter calibration from a single image taken in any position increase efficiency. In clinical settings, this alternate technique allows for rapid recalibration if the focal length is changed, decreasing the total time required to adjust the surgical equipment.

In 2023, Ansari et al. introduced the VentroAR platform for ventriculostomy, with metrics showing a mean (SD) targeting accuracy of 10.64 (5.0) mm, which was measured in terms of the distance of the user's hitting point to the target or the total error. The mean (SD) targeting accuracy of the free hand group after training was 16.7 (5.4) mm in a 2021 study by Van Gestel et al. (39). In a 2022 study by Steiert et al., AR-assisted cranioplasty was employed for single-step resection and cranioplasty, although detailed quantitative results were sparse (21). In a 2022 study, Pojskic et al. combined AR with intraoperative computed tomography navigation for meningioma resections, showing promising outcomes in terms of neuronavigation accuracy and resection success (20). Gross total resection was achieved in 26 of 39 (66.7%) patients, a rate the authors describe as consistent with the published data for similar skull base meningiomas. The accuracy of AR systems is often quantified using target registration error (TRE), which measures the alignment between projected and actual anatomical structures. Automatic registration applying intraoperative computed tomography resulted in high accuracy in registering a target point to a model, which had a mean (SD) TRE of 0.82 (0.37) mm. This measurement is dramatically lower than landmark-based neuronavigation, which had a mean (SD) TRE of 1.85 (1.02) mm, as shown in the study by Bopp et al. (40). Additional AR studies described various applications and outcomes, including a study on retrosigmoid craniotomy planning and AR integration with endoscopic views, underscoring AR's diverse applicability in enhancing surgical precision and workflow (19, 22).

AR technologies have demonstrated varying degrees of accuracy depending on the specific applications, with studies reporting TRE as low as 0.82 mm with the use of intraoperative computed tomography. Reported calibration times in AR studies range from 0.11 to 0.94 s. Although to date few studies have reported improved outcomes associated with AR-assisted surgery, it is important to note that surgical outcomes are not consistently reported across all studies. Among the studies in which outcomes were detailed, there were 60 gross total resections (18, 20, 21), 3 near-total resections (18), 15 subtotal resections (18, 20, 21), 10 partial resections (20, 21), and 5 biopsies [including 2 open biopsies (20, 21) and 3 intraoperative biopsies (21)]. Although specific complication rates are reported far less frequently, the general efficacy of AR in improving surgical navigation and outcomes is well-documented. For example, AR-assisted cranioplasty and endoscopic transsphenoidal surgery show promising results in enhancing accuracy and reducing procedural errors (18).

However, this advancement in AR-guided neuronavigation comes with an added financial burden to the healthcare facility. A study by Davis et al. showed that the total cost of setting up Virtual Interactive Presence in Augmented Reality for a year would be $14,930.39 (41). The Vuzix smart spectacles and the Proximie system retail for $17,000 and $1,300–$2,500 annually, respectively (42). AR software licenses cost $7,000, and hardware costs $2,200, based on the study by Vyas et al. (43). Even though these rates are specific for each region, the cost of a number of these characteristics differs based on the location and the limited supply of goods and services in the area.

A 2024 study by Marrone et al. compared standard magnetic neuronavigation (Medtronic Stealth Station S8) with MR neuronavigation under different lighting conditions (30). Although detailed quantitative results were not provided, MR navigation was noted to enhance visualization and potentially improve surgical accuracy. In another study, Eom et al. investigated MR-guided external ventricular drain placement, in which user trials consisted of 1 blind and 1 MR-assisted drilling of a burr hole followed by a routine, unguided external ventricular drain catheter placement for each of 2 different drill bit sizes. The mean (SD) distance from the catheter target improved from 18.6 (12.5) mm to 12.7 (11.3) mm (p < 0.001) using MR guidance for trials with a large drill bit and from 19.3 (12.7) mm to 10.1 (8.4) mm with a small drill bit (p < 0.001) (25). This study found that real-time quantitative and visual feedback of an MR-guided burr hole procedure can independently improve procedural accuracy. Zeiger et al. reported the utility of MR in various endoscopic skull base procedures, including pituitary tumors and anterior skull base meningiomas, in a cohort of 134 patients (35). Surgeons in this series reported that the technology was particularly useful in less commonly used surgical approaches, such as transorbital surgery. The series reported no intraoperative complications, although some patients did report postoperative complications following surgery of the anterior skull base (35). A 2018 study by McJunkin et al. focused on the use of MR for lateral skull base anatomy, emphasizing the potential of MR to enhance anatomical understanding and precision in temporal bone surgery (31).

The mean (SD) registration error reported in all included MR studies is 5.76 (0.54) mm (35). It is usually measured as the average distance between the virtual object placed in the real world and its intended position, often using specific markers or reference points. In a series of 118 tumor surgeries using MR, reported outcomes included a postoperative death due to respiratory failure (1.5%), hospital readmission (6.7%), and other issues such as cerebrospinal fluid leaks and visual defects, not necessarily due to the use of MR itself. In these cases, resection outcomes varied, with gross total resection achieved in approximately 56.3% of cases and subtotal resection achieved in 43.7% of cases (35). Despite some variability in procedural details and accuracy metrics, MR demonstrates promising potential for enhancing surgical precision and improving patient outcomes.

VR technology has been used for preoperative planning and intraoperative guidance in skull base surgery. A study by Filimonov et al. explored VR applications for endoscopic transnasal approaches to the craniovertebral junction, noting that VR systems provided valuable preoperative insights, including safe surgical planning and visualization of critical structures, despite the small sample size (n = 5) (26). A 2020 study by Liu and Yi identified a new bony landmark (point O, which is the junction point of the temporosphenoid suture and the infratemporal ridge) for lateral skull base surgery using VR, focusing on anatomical precision in cadaveric studies, and the distances measured were in agreement with the actual distances obtained by cadaveric dissections. The mean (SD) dissection vs. mean (SD) VR 3D stereoscopic image was 22.52 (2.47) vs. 22.42 (2.85) for point O to the foramen rotundum (p = 0.81), 22.62 (2.60) vs. 23.51 (2.09) for point O to the foramen spinosum (p = 0.67), 23.69 (2.34) vs. 22.77 (2.90) for point O to the foramen ovale (p = 0.70), and 24.42 (2.38) vs. 24.19 (2.65) for point O to the superior orbital fissure (p = 0.59) (29). The early feasibility study by Won et al. highlighted VR's potential in rehearsing complex endoscopic skull base procedures, with a reported simulation time between 1 and 2 h depending on the number and complexity of the segmented structures, concurrently allowing the surgeon to begin a virtual dissection of the sinuses and ventral skull base (34). In a 2006 study, Rosahl et al. demonstrated the application of VR in enhancing the surgical field through image guidance, although detailed statistics on the impact of VR on surgical outcomes were not extensively documented (33).

The average computation times for VR simulations ranged from 15 s to 2 min, reflecting variability based on the surgeon's experience in this emerging modality. Complications reported in these studies were not due to the VR systems themselves but included intraoperative cerebrospinal leaks and prolonged intubation, although detailed rates were not uniformly provided. Although VR studies do not consistently report specific accuracy metrics, the role of VR in enhancing surgical planning and visualization is evident.

In training environments, high-fidelity VR systems enable surgeons to practice procedures on virtual patients, providing scalable and cost-effective alternatives to traditional cadaveric models (53). These fully digitized surgeries also introduce new methods for assessing surgical skills by capturing performance metrics that are challenging to measure in real-world scenarios (54).

The comparative analysis of XR, based on the limited literature available, found that the pooled mean registration error was 5.76 mm for MR, 6.0 mm for VR, and 6.5 mm for AR (31). Integrating MR, VR, and AR technologies into skull base surgery aims to significantly enhance surgical precision and planning, with no evidence suggestive of increased complications. Overall, MR technology was found to offer improved registration accuracy and resection outcomes, VR technology was found to enhance preoperative planning and simulation, and AR technology was found to provide real-time neuronavigation improvements. The role of XR as an educational tool in skull base surgery is summarized in Table 5 (39, 44, 46–52, 54–56, 61).

Despite their transformative potential, XR applications in skull base surgery face several limitations. “Crowding of objects” in digital overlays can overwhelm surgeons and reality algorithms, necessitating interfaces that prioritize critical information to reduce cognitive overload (112). Additionally, head-mounted devices remain cumbersome, causing physical fatigue during extended procedures, which could potentially reduce accuracy and negatively affect outcomes; lighter ergonomic designs and alternative display technologies, such as integrated operating room systems, are needed (113, 114). Oculomotor abnormalities, nausea, disorientation, and discomfort were among the common negative effects of virtual reality shown by a recent systematic study by Cossio et al. (115). Furthermore, current magnetic resonance scanning resolution limits accurate anatomical segmentation, although advancements in artificial intelligence and high-resolution imaging could address these challenges (116). Ensuring precise overlays demands extensive hardware-software integration, with ongoing efforts required to enhance system reliability, minimize latency, and improve calibration (117, 118). Data security, data management, and the potential for attention bias further underscore the importance of tailored interfaces and robust encryption methods (110). The cost associated with establishing these technologies also poses a significant challenge, especially in hospitals in LMICs.