The application of augmented reality technology in endoscopic pituitary adenoma surgery via nasal approach

Introduction

Pituitary adenoma (PA) is a slow-growing tumor originating from the anterior pituitary, accounting for approximately 10–15% of all intracranial neoplasms (1). Although most PAs are benign, they impose a significant burden on both patients and healthcare systems. The main therapeutic options include pharmacological management, surgical excision, and radiotherapy, with transnasal surgical resection emerging as the primary treatment modality, particularly for alleviating symptoms such as visual impairment caused by the tumor.

In recent years, transsphenoidal endoscopic surgery has become a standard approach for managing a variety of lesions involving the pituitary gland and anterior cranial base (2). By accessing the pituitary gland through the nasal cavity and sphenoid sinus, endoscopy obviates the need for brain retraction, thereby minimizing patient trauma. The close-up view offered by endoscopy allows surgeons to examine the surgical field from within the anatomical structures, providing a broader and more detailed perspective (3). This has led to improved tumor resection rates (4). Moreover, image-based optical neuronavigation enhances intraoperative localization accuracy, reducing the risk of neurovascular injury and improving surgical safety. However, optical neuronavigation systems have notable limitations, such as reliance on two-dimensional (2D) imagery lacking depth perception, high costs, and the necessity of maintaining strict positional stability between the patient and registration reference points.

The advent of augmented reality (AR) technology offers a potential solution to these limitations. AR integrates three-dimensional (3D) imaging into the intraoperative environment, providing enhanced visualization and orientation. Leveraging 3D-Slicer (https://www.slicer.org/) for reconstructing sellar structures has already proven effective for guiding surgical orientation and preoperative planning (5,6). However, despite its promising potential in neurosurgical domains such as transnasal endoscopic surgery, the widespread clinical adoption of AR technology faces significant hurdles. Key challenges include:

The preoperative planning workflow for existing AR systems (involving image segmentation and pathway design) is often time-consuming (7) and requires specialized hardware (e.g., optical tracking equipment), demanding considerable technical training for surgical teams (8).

High initial investment and maintenance costs associated with AR systems, which rely on advanced imaging equipment, proprietary navigation software, and customized hardware (e.g., electromagnetic tracking probes), often exceed those of traditional neuronavigation, hindering adoption in resource-limited settings (9).

Achieving consistently sub-millimeter target registration error (TRE) for intraoperative application remains difficult, and reported TRE values in existing studies vary considerably (10).

The application of AR systems in surgery is further constrained by hardware and software limitations. The heterogeneous nature of different AR solutions (utilizing optical tracking, electromagnetic tracking, or markerless approaches; employing endoscopic overlays or head-mounted displays) leads to a lack of standardization in data formats, registration algorithms, and user interfaces. This fragmentation impedes collaborative multi-center research and technological refinement. Additionally, concerns exist regarding potential visual interference and cognitive overload caused by AR overlays, such as “attentional tunneling” (where surgeons over-rely on virtual cues at the expense of actual anatomical details) or obscuration of subtle tissue changes like minor bleeding (11).

Building upon previous findings, this study developed a novel system integrating AR technology with neuroendoscopy to enhance preoperative planning and intraoperative lesion localization for transnasal PA surgery. Leveraging the powerful modeling capabilities of the widely adopted open-source software 3D-Slicer, we established a cost-effective, user-friendly AR solution with minimal hardware requirements, designed to meet the clinical needs for precise and safe surgical navigation. The system provides AR visualization of critical anatomical structures (e.g., optic nerves, internal carotid arteries, and tumors). Furthermore, its accuracy and safety for neuroendoscopic procedures were rigorously validated through both 3D-printed resin models and clinical cases. We anticipate that the deep integration of AR with neuroendoscopy will significantly improve intraoperative localization efficacy, optimize the operator’s experience, reduce the risk of postoperative complications, and ultimately enhance surgical safety and efficiency, thereby laying the foundation for next-generation intelligent surgical navigation systems. We present this article in accordance with the SUPER reporting checklist (available at https://gs.amegroups.com/article/view/10.21037/gs-2025-95/rc).

Methods

Integrating AR with the endoscopic display involves the following steps (Figure 1):

• Acquisition and importation of imaging data.
• 3D reconstruction.
• Design of the neuroendoscopic guide plate and the endoscopic pathway, followed by 3D printing of the guide plate.
• Integration of endoscopic video into 3D-Slicer.
• Registration of AR models with the endoscopic view.

Figure 1 The complete workflow of AR navigation. 3D-Slicer (https://www.slicer.org/). 3D, three-dimensional; AR, augmented reality; DICOM, Digital Imaging and Communications in Medicine.

Software and hardware

3D-Slicer (version 5.6.1) was used for modeling, while the PLUS (Public software Library for UltraSound imaging research) toolkit integrated endoscopic images into 3D-Slicer. PLUS (version 2.8.0), an open-source toolkit, facilitates data acquisition and calibration from medical devices like optical trackers and surgical navigation systems (12).

The hardware setup included a portable laptop (Y9000p, Lenovo, Beijing, China), a 4K video capture card (4K PRO BOX, Shenzhen IFAN Technology Co., Shenzhen, China) and a KARL STORZ endoscope (Germany).

General information (case details and imaging examinations)

Case information for model validation

Digital Imaging and Communications in Medicine (DICOM) data were collected from five patients diagnosed with PA who underwent transnasal endoscopic surgery at the Department of Neurosurgery, The First Affiliated Hospital of Xiamen University, between March 2021 and April 2022 (Table 1). The cohort included three males and two females, aged 43 to 61 years, with a mean age of 52 years. Among these cases, four were classified as macroadenomas and one as a microadenoma. Postoperative pathological analysis identified two adrenocorticotropic hormone-secreting adenomas, one prolactinoma, and two non-functioning adenomas.

The inclusion criteria encompassed patients clinically diagnosed with PA, aged 18 years or older, with no prior history of transnasal endoscopic surgery, and who had undergone preoperative thin-slice magnetic resonance imaging (MRI) and computed tomography angiography (CTA) scans of the head, while exclusion criteria involved patients with missing preoperative imaging data or incomplete medical records.

Case information for surgical validation

Similar to previous studies (13-16) on AR-assisted surgery that typically adopted small sample sizes (ranging from 3 to 9 cases) to preliminarily assess the technique’s safety, workflow, and potential value, the present study enrolled seven actual surgical cases. Seven patients scheduled for transnasal endoscopic surgery were recruited from the Department of Neurosurgery, The First Affiliated Hospital of Xiamen University, between January and March 2024. The cohort comprised three males and four females, aged 37 to 62 years.

The inclusion criteria comprised patients clinically diagnosed with PA who showed inadequate response to or could not tolerate the adverse effects of standard pharmacological treatment, had no history of transnasal endoscopic surgery, presented no active infections in the nasal cavity, pharynx, or paranasal sinuses affecting the transnasal surgical approach, had clear indications for transnasal surgery, and underwent preoperative MRI and CTA scans of the head. Exclusion criteria included patients lacking definitive preoperative imaging data, those who declined AR-integrated neuroendoscopic assistance, individuals with incomplete medical records, patients with contraindications for surgery, and those requiring emergency intervention (e.g., pituitary apoplexy) precluding elective surgery.

All patients underwent preoperative CTA and MRI examinations. CTA scans were performed using a Siemens 64-slice spiral CT scanner (Siemens Healthcare, Forchheim, Germany) with a slice thickness and interval of 1.00 mm. MRI images were obtained using a 3.0-T MR system (GE Healthcare, Tianjin, China), including T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), contrast-enhanced T1WI and T2WI sequences, with a slice thickness of 1 mm. The following preoperative characteristics were collected and recorded: sex, age, tumor size, and the Knosp classification (based on contrast-enhanced preoperative T1WI) (Table 2). Postoperative data included the extent of tumor resection (assessed via postoperative MRI), operative time, intraoperative blood loss, postoperative hospital stay, and postoperative complications. Observed postoperative complications included—(I) endocrine complications: transient diabetes insipidus, permanent diabetes insipidus, hypopituitarism, and syndrome of inappropriate antidiuretic hormone secretion (SIADH). (II) Neurosurgical complications: intraoperative and postoperative cerebrospinal fluid (CSF) leakage, sellar region hemorrhage, and critical neurovascular injury. (III) Medical complications: meningitis diagnosed via CSF culture, deep vein thrombosis (DVT), and pulmonary embolism.

Ethical approval and patient consent

This study protocol was reviewed and approved by the Institutional Ethics Committee of The First Affiliated Hospital of Xiamen University (No. 2023-091). The study was conducted in strict accordance with the ethical principles outlined in the Declaration of Helsinki and its subsequent amendments.

Model validation section: patient imaging data (DICOM format) used for 3D modeling and printing were retrospectively collected from the hospital’s picture archiving and communication system (PACS). As this component of the study utilized only fully anonymized imaging data for model construction and in vitro validation, entailed no risk of patient identification, and imposed no additional interventions or patient burden, the Ethics Committee approved a waiver of the requirement to obtain specific informed consent for these patients.

Surgical validation section: all patients enrolled in the surgical validation phase were fully informed by the research team prior to surgery regarding the study objectives, application of AR technology, potential risks and benefits, alternative options (traditional navigation), data collection and usage procedures, privacy protection measures, and their right to voluntary participation and withdrawal at any time. Patients who understood and agreed to participate voluntarily provided written informed consent. The informed consent process ensured adequate time for questions and clarifications. All collected clinical data and imaging materials underwent rigorous anonymization to protect patient privacy.

3D modeling process

DICOM data were retrieved from the hospital’s PACS. Using 3D-Slicer software, a 3D model of the sellar region and adjacent structures was created (Figure 2A), highlighting key anatomical features such as the anterior wall of the sphenoid sinus, sphenoid sinus septations, clival recess, sellar floor, optic nerve protuberance, internal carotid artery (ICA) protuberance, optic nerve-internal carotid artery recess (OCR) and tumor, each color-coded for clarity. A tubular model matching the diameter of the endoscope was generated to simulate the optimal surgical pathway, extending from the nostril to the sellar floor (Figure 2B).

Figure 2 Schematic diagram of 3D reconstruction. (A) The completed model includes the tumor (purple), optic nerve (yellow), internal carotid artery (red), sphenoid sinus cavity (brown), and simulated endoscopic pathway (brown rod), displayed in fusion with CTA images in a 3D view. (B) The reconstructed simulated endoscopic pathway (yellow tubular structure), endoscopic guide plate (yellow planar structure), tumor (purple), optic nerve (light yellow), and internal carotid artery (red) are shown in a 3D fusion view. 3D, three-dimensional; CTA, computed tomography angiography.

Neuroendoscopic guide plate design

The neuroendoscopic guide plate ensures the neuroendoscope adheres to the preplanned trajectory, optimizing visualization of anatomical structures. It also limits angle and positional changes, providing stable reference points for AR model registration with the endoscopic field.

The guide plate design process involves modeling the patient’s facial contour, constructing grooves to fix the neuroendoscope along the planned trajectory, and retaining supportive structures such as the forehead, brow arches, and nasal bridge (Figure 3A). Before AR registration, the 3D-printed guide plate is fixed onto the skin, and the neuroendoscope is inserted into the reserved groove and secured in position for subsequent registration (Figure 3B-3D).

Figure 3 Construction and application scenarios of endoscopic guide plate. (A) The reconstructed 3D model of the guide plate. (B) The 3D-printed guide plate. (C) The guide plate, its corresponding 3D-printed patient head model, and the endoscope used. (D) The application scenario of the guide plate. 3D, three-dimensional.

Fabrication of 3D-printed resin model

After 3D modeling, the data were used to 3D-print a photopolymer resin model of the patient’s skull using a 3D printer (Objet350 Connex3, Stratasys, Eden Prairie, USA). The model clearly showed the tumor’s location, size, shape, attachment sites, and its spatial relations with the sellar floor, clival recess, optic nerve, ICA, and cavernous sinus. The model was used to simulate real surgical scenarios, enabling measurement of the registration error under endoscopic guidance to assess the safety and accuracy of the AR system.

Integration of AR with neuroendoscopy

The endoscopic video feed was imported into 3D-Slicer, and AR registration was performed using our self-developed module, “Vrendo”, enabling rapid semi-automated alignment.

The registration process involved:

• Importing data and positioning the guide plate and endoscope.
• Aligning the endoscope path perpendicular to the endoscopic video plane, verified using a region of interest (ROI) framework (Figure 4A,4B).
• Identifying anatomical landmarks (e.g., sphenoid sinus septations, sellar floor, optic nerve protuberance) and selecting reference points. Corresponding points on the 3D model were aligned with these references by translating and scaling the AR model in the endoscopic video plane, ensuring spatial and scale consistency (Figure 4C).

Figure 4 The fusion process of augmented reality and neuroendoscopy. (A) Using the Line tool in the Markups module, a reference line (L) was drawn to represent the endoscopic pathway. The two endpoints of reference line (L) were aligned with the endpoints of the tubular model representing the endoscopic pathway, ensuring that line L passed through the centerline of the endoscope. (B) With the ROI as a reference, the reference line for the endoscopic pathway (red arrow) was adjusted to be perpendicular to the plane of the endoscopic video (blue shaded area). (C) The size and position of the endoscopic view were matched to the 3D model by aligning the corresponding points marked on the 3D model (point F) with the reference points in the endoscopic field of view. (D) The accuracy of the registration was verified using neuronavigation. 3D, three-dimensional; ROI, region of interest.

After AR registration, key anatomical landmarks under the endoscopic view were verified using neuronavigation (Figure 4D). If the AR model’s anatomical localization aligned with the neuronavigation system, the registration was deemed successful. As selecting and aligning reference and corresponding points needed visual confirmation, three different physicians performed AR registration for each case to control potential errors.

Standardized pre-study operational training protocol

Prior to the formal commencement of the study, all three participating neurosurgeons were provided with a standardized preparatory training program accumulating to approximately 20 hours in total. This initiative aimed to ensure operational standardization and data consistency throughout the research. The structured training protocol comprised the following three core components:

3D-Slicer software proficiency training: trainees systematically completed a cumulative 10-hour tutorial-based curriculum utilizing the open-source 3D-Slicer software platform. This module was designed to achieve comprehensive proficiency in software functionalities, with specific emphasis on 3D modeling, segmentation, and measurement operations critical to subsequent research procedures.

Theoretical training on AR registration: each surgeon underwent over 4 hours of systematic instruction on the principles of AR spatial registration. This component ensured a robust conceptual understanding of the core AR co-registration mechanisms, thereby establishing clear expectations regarding the achievable accuracy and efficacy of the AR navigation technology.

Practical AR registration skill acquisition: on tangible 3D-printed anatomical models, each participant independently performed a minimum of 10 AR spatial registration procedures. This hands-on module emphasized the development of practical competencies in achieving precise and efficient spatial alignment between virtual information and physical models within a simulated environment.

Surgical procedure

The surgical team consisted of two senior neurosurgeons specialized in endoscopic skull base surgery (both with >10 years of experience in transnasal pituitary adenomectomy and had independently performed >50 neuroendoscopic procedures), three neurosurgeons responsible for AR registration (all proficient in endoscopic surgical workflows who completed training in 3D-Slicer software operation and AR registration procedures preoperatively, with each having independently performed at least 10 AR registration processes on 3D-printed skull models), one engineer responsible for software technical support, and two specialized nurses familiar with the AR-assisted endoscopic procedure.

Preoperatively, standard neuronavigation registration was performed using the patient’s CTA and MRI data. During surgery, neuronavigation was employed to locate key anatomical structures in the sellar region, such as the sellar floor, ICA protuberance, and sphenoid sinus septations, which served as references for AR registration and validation of registration accuracy.

All procedures were performed in a standard International Organization for Standardization (ISO) Class 5 (cleanroom) operating theater at a tertiary hospital (The First Affiliated Hospital of Xiamen University) via a bilateral transnasal sphenoidal approach. Patients were positioned supine with the head fixed in a Mayfield head holder. The endoscope was introduced through the right nostril, and an arcuate incision was made between the posterior edge of the nasal septum and the inferior turbinate to separate the mucosal flap. A front sphenoidotomy was then completed. The guide plate was positioned, and AR navigation was registered using anatomical landmarks such as the sphenoid sinus septations or clival recess, with neuronavigation used for validation. AR navigation provided a clear understanding of the relative positions of critical structures such as the optic nerve, ICA and tumor, minimizing the risk of damaging these structures during sellar floor opening. A high-speed drill was used to remove the sphenoid sinus septations, and the sellar floor mucosa was displaced laterally. The dura mater was incised in a cruciate fashion to explore the tumor’s size and location, followed by tumor debulking and complete resection. If intraoperative CSF leakage occurred, autologous fascia lata from the thigh was used for repair, along with an absorbable polymer sealing membrane. The mucosal flap was repositioned and secured with biological glue, followed by hemostatic packing.

At the start of surgery, the patient’s anatomy closely matched the 3D model. However, as the endoscope penetrates deeper and surgical manipulations (e.g., drilling sphenoid sinus septations) alter the anatomy, maintaining accurate AR registration becomes more complex. To ensure continuous AR navigation, registration reference points must be dynamically adjusted. Before drilling, the septations or their junction with the sellar floor can serve as reference points. After removal, other high-visibility landmarks, such as the clival recess or optic chiasm prominence, can be used. In summary, dynamically identifying recognizable anatomical structures within the endoscopic view is crucial for maintaining accurate registration.

Statistical analysis

Data are presented as mean ± standard deviation (SD). TRE was calculated across 15 measurements (3 surgeons × 5 models) and reported with a 95% confidence interval (CI) using the t-distribution. Inter-operator variability in TRE was assessed by one-way repeated-measures analysis of variance (ANOVA). Clinical outcomes (e.g., operative time, blood loss) were summarized descriptively for n=7 surgical cases. Analyses used SPSS 27.0 (α=0.05).

Results

3D modeling and AR integration

The application of AR enabled 3D visualization of anatomical structures directly within the endoscopic field. This provided a deeper level of visual insight and enhanced spatial perception (Figure 5).

Figure 5 A 3D modeling and AR integration visualization for one patient. The anatomical structures depicted include the tumor (purple), bilateral optic nerves and optic chiasm (bright yellow), bilateral internal carotid arteries (red), sellar floor bone, and sphenoid sinus septations (desert yellow). (A) A 3D reconstructed model of the sellar region. With the sellar floor bone rendered semi-transparent, the outlines of the tumor, internal carotid arteries, sphenoid sinus cavity, and clival recess are visible. (B) Using 3D-Slicer software, the 3D model was overlaid onto the endoscopic view intraoperatively to achieve augmented reality. The anatomical structures shown include the tumor (purple), bilateral optic nerves and optic chiasm (yellow), bilateral internal carotid arteries (red), and sphenoid sinus cavity (brown). 3D, three-dimensional; AR, augmented reality.

Model validation results

Five patients’ sellar regions were successfully reconstructed into 3D virtual models (Figure 6A), which were subsequently materialized into 3D-printed resin models (Figure 6B). Comparing preoperative 3D reconstructions with intraoperative endoscopic views of key anatomical landmarks showed a high degree of correspondence (Figure 7A-7C). The neuroendoscopic guide plate enabled accurate centering and illumination of the sellar floor (Figure 7D), while AR objects were successfully registered with the endoscopic field, demonstrating high registration accuracy with no visible misalignment (Figure 7E).

Figure 6 Construction of a 3D-printed model of the patient’s sellar floor for model validation. (A) A 3D reconstructed model of the sellar region’s anatomical structures, including the simulated endoscopic pathway (yellow rod). (B) The photopolymer resin model of the skull and sellar region, designed using 3D modeling. 3D, three-dimensional.

Figure 7 Schematic diagram of the model validation results for one patient. (A) Intraoperative endoscopic view. (B) Simulated endoscopic view from the 3D virtual model. (C) 3D reconstructed model of the sellar region. (D) Simulated endoscopic view using the 3D-printed photopolymer resin model. (E) Integrated display of all anatomical structures in the endoscopic view, including the tumor (purple), optic nerve (yellow), and internal carotid artery (red). Reference line L measures 2 mm in length. (F) Schematic of the registration error measurement method, with the AR registration error measured at the cavernous segment of the internal carotid artery (reference line L1) being approximately 1.36 mm. (E,F) It is evident that AR navigation intuitively displays the relative positions of anatomical structures in the sellar region within the endoscopic field. Furthermore, the registration accuracy between the virtual model and the physical model is satisfactory. 3D, three-dimensional; AR, augmented reality.

TRE was measured by comparing the projected contours of the ICA (the cavernous sinus anterior bend segment) in the AR models with their real contours in the endoscopic field (Figure 7F). Across 15 registration attempts (performed by three independent surgeons on five cases), the TRE ranged from 1.36 to 3.45 mm, with an average TRE of 2.23±0.57 mm (Figure 8). A 95% CI for the true population mean TRE (µ), calculated using the t-distribution to account for the sample size (n=15) and unknown population SD, was (1.91, 2.54) mm. Repeated-measures ANOVA revealed no statistically significant differences in TRE among the three operators (P>0.05). These results met the precision requirements for intraoperative navigation.

Figure 8 TREs of five model validation patients (measured at the cavernous segment of the internal carotid artery). The TREs of all 15 registration procedures were within 3.45 mm. Of these, 13 registration procedures had TREs less than 2.67 mm, accounting for 86.7% of the total, while only two procedures had TREs greater than 3.00 mm. Six procedures had TREs below the expected value of 2.00 mm, representing 40.0% of the total. The average TREs for all five cases was within 2.63 mm, which generally met the precision requirements for intraoperative navigation. TRE, target registration error.

The results of the model validation preliminarily demonstrated the reliability of the AR technology developed in this study when integrated with endoscopy. However, it must be noted that all TREs were derived from 3D-printed resin models, and the performance of this AR-endoscopy system may vary in real patients. Therefore, to further evaluate the intraoperative guidance capability of this AR technology, an additional seven patients were recruited for surgical validation.

Clinical validation in real patients

In seven surgical cases, AR models effectively outlined the tumor’s shape, size, and location, as well as adjacent structures like the ICA and sellar floor (Figure 9A,9B). The preplanned trajectory guided the endoscope to the sellar floor’s center, with AR models precisely aligning with intraoperative anatomy. The AR system integrated smoothly with optical neuronavigation and standard neuroendoscopy, offering real-time displays during surgery (Figure 9C).

Figure 9 Schematic diagram of the clinical validation results for one patient. (A) Endoscopic view. (B) AR navigation view. The AR model is aligned with the endoscopic view using the clival recess (brown) and tumor (purple) as reference points, revealing the relative positions of the internal carotid artery (red) and optic nerve (yellow). (C) Intraoperative parallel display: the left monitor shows neuronavigation, the central monitor displays the neuroendoscopic view, and the right monitor presents AR navigation. AR, augmented reality.

Grinding away the sellar floor to expose the underlying PA is a critical step in endoscopic surgery that significantly influences the extent of tumor resection. It is also the stage most prone to intraoperative complications, particularly injury to the ICA. Determining the optimal extent of bony removal requires a comprehensive assessment of tumor size, location, and invasion pattern, balanced against the anticipated difficulty of skull base reconstruction.

AR integration enhances the surgeon’s confidence during this stage. It reduces the difficulty and time required for anatomical identification, thereby decreasing the probability of injuring critical neurovascular structures when opening the sella. For instance, when encountering scenarios like that depicted in Figure 9A—where anatomical landmarks [such as the carotid prominence (CP), OCR, and optic protuberance (OP)] are poorly defined due to anatomical variation or obscured by intraoperative bleeding—the surgeon faces significant challenges in immediately visualizing the precise course of the ICA and optic nerve beneath the bone, as well as the exact tumor location.

Traditionally, neuronavigation systems were employed, requiring point-to-point correlation with preoperative CTA or MRI scans to identify these structures. This approach is not only time-consuming and labor-intensive, causing frequent interruptions to the surgical workflow, but also places considerable demands on the surgeon’s spatial reasoning ability.

In contrast, AR directly integrated into the endoscopic field of view (Figure 9B) provides the surgeon with an intuitive understanding of the spatial relationships between the optic nerve, ICA, and tumor. Consequently, this facilitates a more precise and confident determination of the required extent of sellar floor removal (Video 1).

AR navigation enhances the visualization of critical structures such as the optic nerve, carotid artery and tumor, minimizing the risk of damage during sellar floor opening. It aids surgeons in accurately identifying and locating key anatomical landmarks, improving tumor resection precision. By deepening spatial understanding and enhancing 3D immersion, AR enriches the endoscopic experience. Comparative validation with traditional optical neuronavigation confirmed AR’s accuracy in all cases.

Surgical outcomes

Among the seven PA patients who underwent surgery using AR endoscopic technology, all had well-pneumatized sphenoid sinuses, and their Knosp grades ranged from III to IV. One patient exhibited optic nerve compression, two had prolactinomas, two had growth hormone-secreting adenomas, and three had non-functioning adenomas. All patients presented with varying degrees of sellar floor expansion. The postoperative hospital stay ranged from 7 to 14 days, with an average of approximately 10 days. Surgical durations were between 3.0 and 4.5 hours (the operative time is defined as the duration from the introduction of the endoscope into the nasal cavity until the withdrawal of the endoscope and completion of nasal packing), and intraoperative blood loss ranged from 90 to 280 mL.

All surgeries were successfully completed as planned. Intraoperative CSF leaks occurred in two cases, which were immediately repaired with autologous fascia lata. Two patients experienced transient diabetes insipidus postoperatively, both of whom recovered after one week of conservative treatment. No significant postoperative complications were observed in the remaining patients (Table 3).

Follow-up MRI 3 months post-surgery revealed satisfactory tumor resection outcomes in all patients (Figure 10). After discharge, all seven patients were followed for 3 months in outpatient clinics, and all showed good postoperative recovery with no newly observed complications.

Figure 10 A comparison of preoperative MRI and follow-up MRI at three months post-surgery in two surgical validation patients demonstrated satisfactory tumor resection outcomes. (A) A 59-year-old male diagnosed preoperatively with non-functioning pituitary adenoma, Knosp grade IV on the left and II on the right. Postoperative imaging showed near-total tumor resection with minor residual tissue in the left cavernous sinus. (B) A 37-year-old female diagnosed preoperatively with prolactinoma, Knosp grade III on the left and I on the right. Postoperative imaging also showed near-total tumor resection with minor residual tissue in the left cavernous sinus. MRI, magnetic resonance imaging.

Time consumption in AR navigation

This study documented the temporal requirements for 3D modeling and AR registration procedures performed by a single neurosurgeon. The 3D modeling time was defined as the total duration encompassing the importation of patient imaging data into 3D-Slicer software, completion of 3D reconstruction of the sella turcica, planning of the endoscopic surgical approach, and design of the guide plate. For the cohort comprising five model validation cases and seven surgical validation cases, the mean modeling time across these 12 procedures was 75.42±11.32 minutes. The operational time was defined as the total duration commencing upon placement of the guide plate and endoscope, encompassing the integration of endoscopic video signals into 3D-Slicer software, execution of AR registration, and culminating in the successful output of images to the operating room monitor. The mean operational time for all 12 cases was 5.63±0.85 minutes.

Discussion

Neuronavigation is widely used in transnasal surgeries. However, there are several limitations:

• The high cost of neuronavigation systems, which require expensive equipment, software, and professional training.
• The navigation display is independent of the endoscopic monitor (17), making it difficult to integrate anatomical information directly into the surgical field, potentially distracting the surgeon (18).
• During surgery, the head frame and reference points of optical navigation must remain fixed. Any displacement of the head frame can cause errors, requiring re-registration.
• Significant disparities exist in the adoption rates and application contexts of neuronavigation technology across different geographical regions (19,20). These variations are primarily attributable to multifaceted factors, including economic development levels, allocation of healthcare resources, policy support, and technological infrastructure. Furthermore, the urban-rural divide within regions exacerbates the inequitable distribution of this technology (21).

AR can utilize mobile devices such as personal computers (PCs) as alternatives to expensive image-guided systems (22,23), offering significant advantages like lower cost and ease of use. 3D-Slicer provide a cost-effective and user-friendly option for intracranial surgical localization.

Bopp et al. (24) demonstrated the use of AR integrated with traditional navigation, where they manually aligned a microscope’s focal point with the patient’s reference array to project tumor and anatomical contours. Inspired by their work, this study combines endoscopic video with virtual anatomical models for AR-based registration.

This study introduces a computer-assisted method to integrate AR with neuroendoscopy for pituitary surgery. We developed a new module, “Vrendo”, in 3D-Slicer to link AR virtual images (such as models of the sphenoid sinus and carotid artery) with the endoscopic field of view. This integration allows for AR surgical navigation directly within the endoscopic perspective, offering an intuitive visualization of anatomical structures without requiring 2D-to-3D conversion.

We measured TRE on resin models of five patients (25,26). A lower TRE indicates better navigation accuracy, with a general target of ≤2 mm for surgical navigation. Furthermore, as noted in studies by Citardi et al. (27,28), the target TRE for next-generation surgical navigation platforms is expected to range between 1.0–1.5 mm, ideally achieving 0.6–1.0 mm. In transnasal pituitary surgery, precise identification of anatomical landmarks is crucial due to the proximity of critical structures like the internal carotid artery and optic nerve. This study set the TRE target at ≤2 mm, in line with expectations for next-generation navigation systems.

The feasibility of AR-assisted endoscopy was validated in models and surgical patients. The 95% confidence interval for the TRE measured on the model in this study was (1.96, 2.27) mm. As this interval encompasses the expected value of 2 mm, the results suggest the reliability of AR navigation. While AR navigation currently has slightly higher TRE compared to traditional systems [which can achieve 1.0–2.5 mm (29) TRE and are generally within 2–3 mm (30,31)], it enhances intraoperative orientation by providing clear 3D visualization of hidden anatomical structures behind the sellar floor. AR reduces the need for multiple monitors and minimizes distractions, enabling surgeons to focus more on the procedure. No significant differences were found in tumor resection rates, complication rates, or overall surgical outcomes.

Previous studies used line frameworks or 2D images (32), which were not sufficiently 3D for effective surgical orientation. By contrast, the integration of AR with the endoscopic field provides 3D visual representation. The introduction of 3D models minimizes the risk of unnecessary surgical injury caused by misidentifying anatomical structures, thereby reducing postoperative complications. AR allowed surgeons to comprehensively examine the intricate bony structures of the skull base both preoperatively and intraoperatively. This was particularly beneficial in cases with anatomical challenges such as complex sphenoid sinus septations or anatomical variations in the internal carotid artery (33).

AR navigation offers a significant advantage by eliminating the need for strict head fixation, unlike traditional optical navigation, which requires the patient’s head to remain immobile relative to a reference point. Any movement of the head frame in traditional systems can disrupt accuracy or render navigation unusable, often necessitating time-consuming re-registration. In contrast, AR relies on a compact, lightweight endoscopic guide plate, which avoids complex setup and calibration, effectively addressing head frame displacement issues and providing a more efficient, user-friendly alternative.

Unlike microscopes, endoscopes are frequently repositioned during surgery to meet exposure needs, making it difficult to maintain a fixed position or follow a pre-planned pathway. Therefore, we developed an endoscope guide plate that temporarily stabilizes the endoscope during AR registration, ensuring alignment with the pre-designed pathway from the nostril to the sellar floor. This guide plate, which is removed post-registration to avoid obstructing surgery, anchors the endoscope’s pitch and horizontal angles, serving as a bridge between the virtual model and real-world anatomy. Recognizable landmarks, such as the optic nerve protuberance and sphenoid sinus septations, were used to align the endoscopic video with the virtual model. Once registered, less prominent structures like the carotid artery and tumor can be located more precisely. Although the guide plate may become temporarily unusable as the endoscope’s depth and angle change during surgery, the surgeon’s spatial understanding, supported by reference screenshots, ensures confident navigation.

TRE may arise from two sources. First, poor fit between the guide plate and the facial surface or insufficient adherence to the endoscope can be mitigated by slightly increasing the guide plate’s thickness (3–5 mm) and extending the endoscopic channel by 4 cm. Second, inaccuracies in 3D modeling of the endoscopic channel may misdirect the endoscope, causing misalignment with virtual structures. To control this, the endoscopic pathway model should be extended at both ends. If errors occur intraoperatively, they can be corrected using recognizable anatomical structures as references, with manual fine-tuning verified by at least three surgeons.

The AR navigation system demonstrates not only a low initial investment but also significant cost advantages over traditional navigation systems. The core hardware comprises solely a laptop, video capture card, and standalone monitor, with implementation necessitating only supplementary procurement of a 3D printer (market price ~$2,000 USD). This yields a total acquisition cost of approximately $4,200 USD—a negligible expenditure compared to the hundreds of thousands associated with neuronavigation systems. Crucially, the system utilizes open-source software (3D-Slicer), eliminating licensing and maintenance expenses. The cost associated with 3D printing and sterilization per surgical procedure was estimated to be as low as approximately 10 USD. Beyond these, no significant operational costs are incurred.

Operational costs were marginal. Preoperative imaging (cranial CTA/MRI) constituted routine clinical workflows, incurring no additional radiation exposure or expenditures. Both 3D-printing costs (encompassing materials, labor, and energy) and sterilization were negligible relative to overall surgical expenses. Moreover, the system leverages existing hospital neuroendoscopes and personal laptops, enhancing feasibility for primary healthcare settings lacking advanced navigation infrastructure.

The open-source 3D-Slicer software, requiring minimal hardware, is user-friendly. By using DICOM data, anatomical structures can be easily reconstructed. The modeling process is simple and can be quickly mastered by neurosurgeons with basic training.

Utilizing the methodology presented in this study, analogous AR overlays can be achieved in surgeries at other anatomical sites employing alternative intraoperative imaging devices (e.g., microscopes, laparoscopes, thoracoscopes). However, this implementation would require either the redesign of the guide plates, the adoption of functionally analogous supportive instruments, or the utilization of specialized surface fiducial markers to provide reference points for registration.

In summary, this study explored AR in transnasal PA surgery, integrating AR with endoscopy using 3D-Slicer. The endoscopic view illuminates the sellar floor, while AR overlays critical structures like the tumor, carotid artery, and optic nerve onto the endoscopic view. This bridges the virtual and real-world environments, providing a safe, convenient, and effective surgical tool for PA treatment.

Limitations

While this study demonstrates the feasibility of AR-assisted neuroendoscopy in transnasal PA surgery, several limitations warrant consideration. Preoperative 3D modeling and planning remain time-consuming, necessitating workflow optimization for clinical scalability. Although the average TRE met feasibility thresholds, achieving consistent sub-2 mm accuracy is challenging. Critically, the static model-based TRE likely underestimates intraoperative error due to tissue shift from drilling/tumor removal, subjectivity in landmark selection under bleeding/distortion, and the inability of rigid models to replicate live tissue compliance and physiological motion. This is clinically significant given the proximity of neurovascular structures (e.g., 3–5 mm optic nerves), where even the observed TRE range (up to 3.45 mm) erodes safety margins, mandating surgical caution beyond AR-indicated boundaries. The system further lacks real-time tracking and dynamic model updates during tissue deformation, requiring manual re-registration. Finally, while AR enhances spatial orientation, it carries a risk of attentional tunneling—over-reliance on virtual overlays may obscure subtle real-time anatomical cues (e.g., micro-bleeding, CSF leaks), emphasizing that AR must serve as an adjunct to surgical expertise and direct observation.

Specific recommendations for future research

Future research should prioritize the integration of computer vision algorithms to enable real-time 3D model updates during endoscopic dynamic motion, thereby eliminating the need for intraoperative manual re-registration. Concurrently, there is a need to develop intelligent registration architectures based on deep learning (e.g., point cloud matching networks), continuing the optimization of TRE. To establish the clinical translational value of AR, prospective randomized controlled trials (RCTs) with large, multi-center samples should be conducted. Furthermore, developing cloud-based platforms for automated 3D model generation would significantly optimize preoperative workflow timeliness and is crucial for the broader adoption of AR technology.

Conclusions

Surgical validations and surgeon feedback show that the AR endoscopic system significantly enhances surgical orientation, saves time and effort, and improves surgeons’ comfort. AR reduces the risk of damaging critical neurovascular structures and enables more precise minimally invasive surgery by visualizing deep anatomical structures that are invisible to the naked eye. Future efforts should focus on optimizing the registration algorithm to minimize errors and integrating a spatial coordinate mapping system to enable free endoscope rotation and real-time AR model tracking during surgery.

Acknowledgments

We sincerely appreciate the technical support provided by The First Affiliated Hospital of Xiamen University.

Footnote

Reporting Checklist: The authors have completed the SUPER reporting checklist. Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-95/rc

Data Sharing Statement: Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-95/dss

Peer Review File: Available at https://gs.amegroups.com/article/view/10.21037/gs-2025-95/prf

Funding: This work was supported by The Clinical Research Fund initiated by the researchers of The First Affiliated Hospital of Xiamen University (No. XMFHIIT-2023SL066). The article processing charge (APC) was funded by The Clinical Research Fund initiated by the researchers of The First Affiliated Hospital of Xiamen University.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://gs.amegroups.com/article/view/10.21037/gs-2025-95/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Ethics Committee of The First Affiliated Hospital of Xiamen University (No. 2023-091) and informed consent was obtained from all individual participants.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Karimian-Jazi K. Pituitary gland tumors. Radiologe 2019;59:982-91. [Crossref] [PubMed]
2. de Divitiis E, Laws ER, Giani U, et al. The current status of endoscopy in transsphenoidal surgery: an international survey. World Neurosurg 2015;83:447-54. [Crossref] [PubMed]
3. Cavallo LM, Dal Fabbro M, Jalalod'din H, et al. Endoscopic endonasal transsphenoidal surgery. Before scrubbing in: tips and tricks. Surg Neurol 2007;67:342-7. [Crossref] [PubMed]
4. Guo-Dong H, Tao J, Ji-Hu Y, et al. Endoscopic Versus Microscopic Transsphenoidal Surgery for Pituitary Tumors. J Craniofac Surg 2016;27:e648-55. [Crossref] [PubMed]
5. Liu J, Wang B. Application of 3D-Slicer Software in the Treatment of Gliomas. J Craniofac Surg 2024; Epub ahead of print. [Crossref]
6. Liu J, Li X, Leng X, et al. Effect of 3D Slicer Preoperative Planning and Intraoperative Guidance with Mobile Phone Virtual Reality Technology on Brain Glioma Surgery. Contrast Media Mol Imaging 2022;2022:9627663. [Crossref] [PubMed]
7. Okamoto T, Onda S, Matsumoto M, et al. Utility of augmented reality system in hepatobiliary surgery. J Hepatobiliary Pancreat Sci 2013;20:249-53. [Crossref] [PubMed]
8. Linxweiler M, Pillong L, Kopanja D, et al. Augmented reality-enhanced navigation in endoscopic sinus surgery: A prospective, randomized, controlled clinical trial. Laryngoscope Investig Otolaryngol 2020;5:621-9. [Crossref] [PubMed]
9. Meola A, Cutolo F, Carbone M, et al. Augmented reality in neurosurgery: a systematic review. Neurosurg Rev 2017;40:537-48. [Crossref] [PubMed]
10. Cui Y, Zhou Y, Zhang H, et al. Application of Glasses-Free Augmented Reality Localization in Neurosurgery. World Neurosurg 2023;180:e296-301. [Crossref] [PubMed]
11. Marcus HJ, Pratt P, Hughes-Hallett A, et al. Comparative effectiveness and safety of image guidance systems in neurosurgery: a preclinical randomized study. J Neurosurg 2015;123:307-13. [Crossref] [PubMed]
12. Lasso A, Heffter T, Rankin A, et al. PLUS: open-source toolkit for ultrasound-guided intervention systems. IEEE Trans Biomed Eng 2014;61:2527-37. [Crossref] [PubMed]
13. Panesar SS, Magnetta M, Mukherjee D, et al. Patient-specific 3-dimensionally printed models for neurosurgical planning and education. Neurosurg Focus 2019;47:E12. [Crossref] [PubMed]
14. Luo H, Yin D, Zhang S, et al. Augmented reality navigation for liver resection with a stereoscopic laparoscope. Comput Methods Programs Biomed 2020;187:105099. [Crossref] [PubMed]
15. Oh MY, Yoon KC, Hyeon S, et al. Navigating the Future of 3D Laparoscopic Liver Surgeries: Visualization of Internal Anatomy on Laparoscopic Images With Augmented Reality. Surg Laparosc Endosc Percutan Tech 2024;34:459-65. [Crossref] [PubMed]
16. Inoue D, Cho B, Mori M, et al. Preliminary study on the clinical application of augmented reality neuronavigation. J Neurol Surg A Cent Eur Neurosurg 2013;74:71-6. [Crossref] [PubMed]
17. Khan R, Plahouras J, Johnston BC, et al. Virtual reality simulation training in endoscopy: a Cochrane review and meta-analysis. Endoscopy 2019;51:653-64. [Crossref] [PubMed]
18. Cho J, Rahimpour S, Cutler A, et al. Enhancing Reality: A Systematic Review of Augmented Reality in Neuronavigation and Education. World Neurosurg 2020;139:186-95. [Crossref] [PubMed]
19. Májovský M, Grotenhuis A, Foroglou N, et al. What is the current clinical practice in pituitary adenoma surgery in Europe? European Pituitary Adenoma Surgery Survey (EU-PASS) results-technical part. Neurosurg Rev 2022;45:831-41. [Crossref] [PubMed]
20. Ademe BW, Tebeje B, Molla A. Availability and utilization of medical devices in Jimma zone hospitals, Southwest Ethiopia: a case study. BMC Health Serv Res 2016;16:287. [Crossref] [PubMed]
21. Xu W, Pan Z, Zhang L, et al. Optimizing the medical equipment investment in primary care centres in rural China: evidence from a panel threshold model. BMC Health Serv Res 2024;24:160. [Crossref] [PubMed]
22. Kramers M, Armstrong R, Bakhshmand SM, et al. Evaluation of a mobile augmented reality application for image guidance of neurosurgical interventions. Stud Health Technol Inform 2014;196:204-8.
23. Hou Y, Ma L, Zhu R, et al. iPhone-Assisted Augmented Reality Localization of Basal Ganglia Hypertensive Hematoma. World Neurosurg 2016;94:480-92. [Crossref] [PubMed]
24. Bopp MHA, Saß B, Pojskić M, et al. Use of Neuronavigation and Augmented Reality in Transsphenoidal Pituitary Adenoma Surgery. J Clin Med 2022;11:5590. [Crossref] [PubMed]
25. West JB, Fitzpatrick JM, Toms SA, et al. Fiducial point placement and the accuracy of point-based, rigid body registration. Neurosurgery 2001;48:810-6; discussion 816-7. [Crossref] [PubMed]
26. Labadie RF, Davis BM, Fitzpatrick JM. Image-guided surgery: what is the accuracy? Curr Opin Otolaryngol Head Neck Surg 2005;13:27-31. [Crossref] [PubMed]
27. Citardi MJ, Batra PS. Intraoperative surgical navigation for endoscopic sinus surgery: rationale and indications. Curr Opin Otolaryngol Head Neck Surg 2007;15:23-7. [Crossref] [PubMed]
28. Citardi MJ, Yao W, Luong A. Next-Generation Surgical Navigation Systems in Sinus and Skull Base Surgery. Otolaryngol Clin North Am 2017;50:617-32. [Crossref] [PubMed]
29. Seno S, Suzuki M, Sakurai H, et al. Image-guided endoscopic sinus surgery: a comparison of two navigation systems. Nihon Jibiinkoka Gakkai Kaiho 2005;108:1101-9. [Crossref] [PubMed]
30. Orringer DA, Golby A, Jolesz F. Neuronavigation in the surgical management of brain tumors: current and future trends. Expert Rev Med Devices 2012;9:491-500. [Crossref] [PubMed]
31. Burduk PK, Dalke K, Kaźmierczak W. Intraoperative navigation system in endoscopic sinus surgery. Otolaryngol Pol 2012;66:36-9. [Crossref] [PubMed]
32. Kawamata T, Iseki H, Shibasaki T, et al. Endoscopic augmented reality navigation system for endonasal transsphenoidal surgery to treat pituitary tumors: technical note. Neurosurgery 2002;50:1393-7. [Crossref] [PubMed]
33. Poirier J, Duggal N, Lee D, et al. Sphenoid sinus septations: unpredictable anatomic landmarks in endoscopic pituitary surgery. J Otolaryngol Head Neck Surg 2011;40:489-92.