Facial Nerve Tractography of Vestibular Schwannomas: A Systematic Review of MR Acquisition and Analysis Pipelines

ABSTRACT

Background

Facial nerve (CN VII) diffusion MR tractography is considered as a useful adjunct in pre‐operative planning prior to vestibular schwannoma (VS) resection, especially in larger (Koos Grade III/IV) tumors. Since 2016, several systematic reviews have investigated the clinical value of CN VII tractography in VS, and all reported a “success rate” of at least 87% for predicting the pre‐operative CN VII position. Yet in clinical practice, CN VII tractography has not yet been widely adopted into routine clinical practice. We suspected that underlying methodology and reporting metrics for existing tractography algorithms may be overestimating success rate. This motivated us to revisit the literature from a different perspective to unravel the caveats and nuances behind this technology.

Methods

We screened all published works on PubMed related to pre‐operative CN VII tractography in VS. Twenty‐two studies were reviewed in detail.

Results

We observed a strikingly high heterogeneity in tractography protocols in all domains of the tractography acquisition and analysis pipeline across studies.

Conclusions

These findings suggest that the reliability and reproducibility of CN VII tractography in large VS has been overestimated. We believe that employing standardized reporting metrics, including sensitivity, true predictive value, and false discovery rate, would increase the transparency of benchmarking over other commonly reported metrics (“success rate” or “concordance rate”). In addition, ongoing research should aim to systematically investigate and improve each step in the acquisition and analysis pipeline for CN VII tractography in VS.

1. Introduction

Vestibular schwannoma (VS) is the most common neoplasm of the cerebellopontine angle (CPA), with an annual incidence of 1.5 to 5.2 per 100,000 people [1, 2, 3]. Depending on tumor size, presenting symptoms, and other patient‐specific variables, treatment may entail microsurgical resection. Current surgical paradigms aim for a maximal safe tumor resection while preserving the integrity and neurological function of the facial nerve (CN VII). Unfortunately, the position and presentation of CN VII in patients with larger (Koos Grade III and IV) VS may in many cases be difficult to predict. This includes both the positional displacement of the nerve relative to the tumor mass (e.g., superior, anterior, inferior) and the fascicular integrity, as fibers are often markedly splayed in large tumors. Given these challenges, finding and preserving CN VII can be time‐consuming, technically difficult, and potentially associated with worsened outcomes.

Due to this phenomenon, there is great value in preoperatively predicting the anatomical course and integrity of CN VII in KOS Grade III/IV tumor presentations. This predictive ability is very limited with conventional magnetic resonance imaging (MRI) techniques (e.g., T₁‐weighted, T₂‐weighted, or steady‐state free precession methods) [4]. In most cases, CN VII is located in the anterior aspect of the tumor [5, 6, 7]. However, this is not always the case and cannot be predicted with certainty on conventional imaging. Since larger tumors are almost always approached from the posterior aspect (via translabyrinthine or retrosigmoid approaches), surgeons must proceed slowly during the initial stages of tumor debulking and manipulation due to the uncertainty of CN VII position. Once the nerve's course is confirmed, surgical resection can shift to a relatively quicker pace. Thus, preoperative knowledge of CN VII anatomy in large VS cases has the potential to increase surgical efficiency and reduce the risk of iatrogenic injury.

Diffusion magnetic resonance (MR) tractography has been recognized as an important preoperative tool for intra‐axial brain surgery, because it can visualize white matter pathways in vivo [8]. Following the Food and Drug Administration (FDA) clearance of MR tractography software on major neuronavigation workstations (i.e., Medtronic, Brainlab), the usage of MR tractography has gradually become a “standard of care” in neurosurgical planning [9, 10]. By delineating the course of major white matter tracts (e.g., corticospinal tract, medial lemniscus) around pathologic lesions, it helps surgeons avoid inadvertent trauma to functional areas of the brain and spinal cord [9, 10].

Given the successes of MR tractography in other domains, there has been growing interest in applying this technology to preoperative mapping of the facial nerve around Koos Grade III/IV tumors [11]. Numerous investigators have subsequently reported on various techniques and protocols for CN VII tractography in AN, achieving variable levels of success. Data from these studies have been summarized across six distinct systematic reviews at the time of this writing [10, 12, 13, 14, 15]. Each of the reviews report an aggregate “success rate” of at least 87% for predicting the preoperative CN VII position with respect to the tumor. However, the protocol validity and clinical reliability of CN VII tractography remains in question. The technique has thus failed to achieve widespread use.

MR tractography of the facial nerve in the setting of VS is among the most technically demanding applications of the technique. This difficulty arises from a convergence of anatomical and methodological challenges. The small diameter of CN VII requires high‐resolution imaging, often at the expense of scan time or signal‐to‐noise ratio (SNR). The proximity to the skull base bone introduces significant susceptibility‐related distortion. Finally, partial volume effects from adjacent cerebrospinal fluid (CSF) and tumor further degrade signal fidelity [16]. These physical constraints are compounded by variability in MR hardware, acquisition parameters, preprocessing steps, signal modeling, and fiber tracking algorithms. While tractography is most desirable for large tumors (Koos Grade III–IV), these cases are also the most limiting for the technology [10, 12, 13, 14, 15, 17]. Collectively, these challenges likely explain the disconnect between the synthesized literature and actual clinical performance. This prompted our team to reinvestigate the variables that most influence outcomes in CN VII tractography.

With this in mind, we sought to comprehensively extract and analyze data from every step (Figure 1) of the tractography pipeline across all studies investigating the preoperative use of CN VII tractography in VS. This work thus aimed to address the limitations inherent to these studies, and to identify critical protocol domains where future research may be of greatest benefit.

FIGURE 1.

CN VII tractography is a multistep process with in. The generation of CN VII tractography usually involves four sequential steps: scanning, preprocessing, signal modeling, and tracking. To obtain raw diffusion image on different MR hardware with different setups, a diffusion MR sequence is performed to scan a patient. Some of the most important parameters of a sequence would include TR/TE, b‐value, gradient encoding directions, and shots. After raw diffusion images are obtained, certain preprocessing steps can improve the imaging quality, such as denoising and motion/artifact correction. This is followed by fitting signal models to “decode” the diffusion signals. The most commonly used model is diffusion tensor, but there has been an increased interest of constrained spherical deconvolution (CSD) because of its superiority in solving crossing fiber issues. Multitissue CSD has been proposed in recent years in attempt to solve the partial volume effect, as it offers signal modeling for different tissue types (e.g., gray matter, white matter, CSF), which is not a feature in single tissue CSD model. After signal modeling, tracking algorithm is deployed to model and generate streamlines across voxels. Deterministic algorithm has been widely used, with emerging algorithms such as probabilistic and geodesic gaining interest in the field. Conventionally, neurosurgeons use Sampath classification system to confirm the relative location of CN VII to tumor by visual confirmation using neuronavigation display in OR. This approach is simple and intuitive. In addition, it is potentially viable to overlay and compare the geometric proximity between the tractography streamlines and real‐time CN VII stimulation points on neuronavigation. However, this methodology has not been reported in the body of literature we reviewed. Abbreviations: CN, cranial nerve; dMR, diffusion magnetic resonance; OR, operating room.

2. Methods

2.1. Study Selection

This systematic review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) guidelines [18] (Figure 2). A literature search was performed in the PubMed (U.S. National Library of Medicine, National Institutes of Health) database. No other databases or backward/forward searching was performed. The search strategy utilized the syntax: “((vestibular schwannoma) OR (acoustic neuroma)) AND (tractography).” The database was searched from January 1, 1990, through December 31, 2024. All references were exported to the Endnote (Clarivate, Philadelphia, PA) citation manager, which were further exported to Microsoft Excel (Microsoft, Redmond, WA) for screening.

FIGURE 2.

The PRISMA flowchart of current study. Abbreviations: n, number; PRISMA, Preferred Reporting Items for Systematic Reviews and Meta‐Analyses. ^*Consider, if feasible to do so, reporting the number of records identified from each database or register searched (rather than the total number across all databases/registers). ^**If automation tools were used, indicate how many records were excluded by a human and how many were excluded by automation tools.

2.2. Inclusion and Exclusion Criteria

Studies were included if they met the following criteria: (1) confirmed diagnosis of VS; (2) in vivo MRI performed; (3) original research studies with a sample size ≥ 5 adult patients undergoing microneurosurgical resection; and (4) validation of preoperative facial nerve MR tractography with intraoperative findings. Exclusion criteria included: non‐English studies, review papers (including systematic reviews and meta‐analyses), studies involving neurofibromatosis‐2 patients, and case series or reports with fewer than five patients.

2.2.1. Screening and Quality Control

Titles and abstracts were screened by two independent reviewers. Full texts were then assessed by one reviewer to confirm eligibility. A senior reviewer conducted a final quality check to validate studies selected for inclusion.

2.2.2. Data Extraction and Classification

We extracted comprehensive data on all relevant variables contributing to CN VII tractography outcomes. As detailed in the following subsections, these were organized into four major categories: (1) MR acquisition parameters; (2) preprocessing and tracking methods; (3) intraoperative validation approaches; and (4) patient demographics and tractography outcomes.

2.2.3. MR Acquisition Parameters

2.2.3.1. Scanner Hardware

We recorded scanner vendor, field strength (1.5T or 3.0T), radiofrequency (RF) coil type, and use of acceleration techniques. Higher field strength offers increased SNR values, enabling shorter and/or higher‐resolution scans, which are crucial for delineating smaller cranial nerves. However, this comes at the cost of increased sensitivity to field variations and associated image distortions. Modern MR scanners also utilize specialized RF coil arrays to enhance SNR and/or enable image reconstruction from undersampled data, the latter of which can also reduce image distortion in diffusion‐weighted data under certain conditions. Finally, multiple scanner vendors (e.g., Philips, Siemens, and GE) are available, each offering proprietary hardware, diffusion MRI (dMRI) sequences, and image reconstruction algorithms, all of which contribute to interscanner variability [19].

2.2.3.2. dMRI Pulse Sequences and Readout Strategies

Diffusion‐weighted single‐shot echo‐planar imaging (EPI) is the standard sequence for dMRI due to its efficiency and insensitivity to bulk patient motion, as all image data are acquired in a single “shot.” However, this approach is prone to distortions in the skull base. Several multishot techniques (e.g., readout‐segmented EPI, RESOLVE) have been employed to reduce these distortions at the cost of decreased efficiency and increased sensitivity to patient motion. In both cases, various undersampling (or acceleration) approaches are also integrated into these sequence readouts, which can impact SNR and image distortions.

2.2.3.3. dMRI Sequence Parameters

First, the strength of the diffusion weighting (or b‐value), along with the number and orientation of diffusion gradient directions, were included in our analyses, as these factors determine the angular resolution and fidelity of fiber orientation estimation in tractography. While higher b‐values enhance sensitivity to water diffusion, they also reduce SNR. Similarly, higher gradient direction sampling can improve our ability to resolve complex fiber geometries, but comes at the cost of increased scan time [8, 20]. These trade‐offs must be carefully balanced when designing protocols for cranial nerve tractography.

In addition, the in‐plane resolution and slice thickness determine the resolution of the scan and our ability to visualize small structures like cranial nerves. Note that while dMRI acquisitions in the brain are generally acquired with isotropic voxels, anisotropic voxels are sometimes used in nerves (due to their elongated structures) to increase SNR.

Finally, sequence timings (repetition time, TR, and echo time, TE) and the number of averages were extracted. Longer TRs can also improve SNR but lead to increased scan times, while shorter TEs can mitigate susceptibility effects but may be impractical on some systems due to hardware limitations. In addition, when diffusion properties differ inside and outside of axons, these parameters can impact downstream fiber‐tracking results [21]. Finally, employing multiple averages boosts SNR at the cost of longer scan times or a reduction in the number of diffusion directions.

2.2.3.4. Structural Imaging Sequence

We extracted the type of structural scan used for registration: T₁‐weighted (T₁W), T₂‐weighted (T₂W), or steady‐state free precession (SSFP). Structural imaging sequences parameters serve as anatomical references to guide region‐of‐interest (ROI) placement for fiber track seed points.

2.2.4. Preprocessing and Tracking Methods

2.2.4.1. Raw Data Display

We critically evaluated whether the raw diffusion MR images were presented, as image quality in the skull base region can often be compromised. Fundamentally, the raw dMR data should at least exhibit some degree of discernible traces of the direction‐dependent diffusion signal that underpins tractography. If these features are absent or unclear in the raw images, it may raise questions about the validity of the reconstructed fibers. Sharing the raw data, therefore, is not only a critical matter of transparency but also essential to instill the confidence of the tractography results.

2.2.4.2. Software

We extracted whether proprietary and third‐party software solutions were used for processing and tractography. Proprietary software is typically integrated within neuronavigation systems, whereas third‐party software is created by independent developers. Often, the algorithm of third‐party software is open source, providing greater transparency.

2.2.4.3. Preprocessing Pipeline

We also extracted details on the preprocessing pipeline, including denoising, motion, and artifact correction methods. Motion correction deals with both patient motion and intrinsic physiological motions (e.g., CSF pulsation or vascular flow). Additionally, eddy current artifacts—arising from the rapidly changing gradient magnetic fields used to encode the diffusion signal—and distortions from magnetic field inhomogeneities at the skull base can also be corrected. While these approaches mitigate image quality issues in the brain, they have not been systematically evaluated for applications in cranial nerves.

2.2.4.4. Signal Modeling and Fiber Tracking Algorithms

We recorded the signal model used to determine the primary direction of nerve fibers within each voxel (e.g., diffusion tensor imaging, or DTI, and constrained spherical deconvolution, or CSD). Whether a single‐tissue or multitissue model was employed (to account for partial volume effects) was also noted. In addition, we evaluated the tractography algorithm used to propagate fiber pathways across voxels, distinguishing between deterministic and probabilistic approaches.

2.2.4.5. ROI Placement for Tracking

We noted ROI selection strategies for fiber tracking, as the quality and consistency of these steps vary across studies. Accurate tractography requires precise ROI placement to constrain tracking to relevant anatomy. For VS, placing ROIs between the internal auditory canal (IAC) and CN VII's root entry zone (REZ) is conceptually straightforward. However, the lower resolution of dMRI compared to structural images can obscure anatomical landmarks. To address this, dMRI is typically registered to high‐resolution structural scans before ROI delineation. The choice of structural sequence and resulting alignment quality can vary.

2.2.4.6. Fiber Filtration Methods

We recorded the fiber filtration strategies used to remove spurious fibers that are common in tractography. Fiber filtering often relies on thresholds for fractional anisotropy (FA) and/or angular deviation to exclude anatomically implausible streamlines. Manual filtering using expert knowledge is also common. In CN VII tractography for VS, filtering may occur in the CPA cistern or brainstem. We specifically examined how studies reported their criteria for retaining or discarding fibers and identifying the true CN VII among spurious tracks.

2.2.4.7. Intraoperative Validation Approaches

The position of CN VII was classified using the Sampath system [7] or its variants. For each study, we assessed the level of detail provided in describing the nerve course (e.g., use of multiple labels or fine‐grained anatomical descriptions) and the method of confirmation (e.g., direct visual inspection or alternative techniques). We also noted whether surgeons were blinded to the tractography data. In addition, we extracted information on intraoperative electromyography (EMG) usage and recorded whether the morphology of CN VII and/or CN VIII (e.g., compact vs. splayed) was described.

2.2.5. Patient Demographics and Tractography Outcomes

2.2.5.1. Patient Demographics

We collected data on sample size, proportion of Koos Grade III–IV cases, and mean tumor diameter as indicators of sample quality (smaller lesions are associated with higher CN VII tractography success).

2.2.5.2. Reported Tracking Outcomes

A common metric used in most studies correlates tractography‐predicted CN VII location with direct intraoperative visualization of the nerve (by surgeons). The percentage of cases where the two align is defined in this study as “concordance rate.” However, in complex scenarios—such as when CN VII is splayed—it may be difficult to determine the nerve's true course during surgery, and the fiber tracking may fail altogether (not “successful”). For each study, we recorded the reported “successful rate” (i.e., if the tractography was successfully rendered) as well as the percentage of cases in which the CN VII/VIII tractography shape corresponded with intraoperative findings.

2.2.5.3. Estimated Tracking Outcomes

For each study, we additionally estimated standard tractography metrics, including true positives (TP), false negatives (FP), and false negatives (FN) [8] defined as:

• True positive (TP): Preoperative CN VII location matches intraoperative findings

• False positive (FP): Predicted CN VII location not identified during surgery

• False negative (FN): Intraoperative CN VII not identified on tractography

From these values, sensitivity (TP/[TP+FN]), positive predictive value (PPV = TP/[TP+FP]), and false discovery rate (FDR = FP/[TP+FP]) were calculated.

2.2.5.4. Statistical Reporting

All categorical variables were summarized as frequencies and percentages, and continuous variables as means with standard deviations (SD). Bias analysis was performed using the QUADAS‐2 protocol [22] by two authors. A meta‐analysis was not conducted due to heterogeneity in study designs and outcome reporting.

3. Results

3.1. Study Selection

A total of 38 articles were identified through the initial PubMed search. After screening titles and abstracts, 16 studies were excluded (eight for incorrect study design [e.g., articles that do not meet our inclusion/exclusion criteria] and eight for article types). Full‐text review excluded one additional study. Ultimately, 22 studies met the inclusion criteria and were analyzed in detail (Table 1).

TABLE 1.

Articles in the final inclusion list.

Year	Journal	First author	Title
2006	Journal of Magnetic Resonance Imaging	Taoka	Displacement of the Facial Nerve Course by Vestibular Schwannoma: Preoperative Visualization Using Diffusion Tensor Tractography
2011	Journal of Neurosurgery	Gerganov	Diffusion Tensor Imaging–Based Fiber Tracking for Prediction of the Position of the Facial Nerve in Relation to Large Vestibular Schwannomas: Clinical Article
2014	Journal of Korean Neurosurgical Society	Choi	Preoperative Identification of Facial Nerve in Vestibular Schwannomas Surgery Using Diffusion Tensor Tractography
2015	Journal of Neurosurgery	Yoshino	Combined Use of Diffusion Tensor Tractography and Multifused Contrast‐Enhanced FIESTA for Predicting Facial and Cochlear Nerve Positions in Relation to Vestibular Schwannoma
2015	Acta Neurochirurgica	Yoshino	Feasibility of Diffusion Tensor Tractography for Preoperative Prediction of the Location of the Facial and VestibuloCochlear Nerves in Relation to Vestibular Schwannoma
2015	Acta Neurochirurgica	Wei	Identification of Cranial Nerves Near Large Vestibular Schwannomas Using Superselective Diffusion Tensor Tractography: Experience With 23 Cases
2016	Journal of Neurosurgery	Song	In Vivo Visualization of the Facial Nerve in Patients With Acoustic Neuroma Using Diffusion Tensor Imaging‐Based Fiber Tracking
2016	Neurology India	Borkar	Prediction of Facial Nerve Position in Large Vestibular Schwannomas Using Diffusion Tensor Imaging Tractography and Its Intraoperative Correlation
2016	Otology & Neurotology	Hilly	Diffusion Tensor Imaging Tractography of the Facial Nerve in Patients With Cerebellopontine Angle Tumors
2017	World Neurosurgery	Li	Identification of the Facial Nerve in Relation to Vestibular Schwannoma Using Preoperative Diffusion Tensor Tractography and Intraoperative Tractography‐Integrated Neuronavigation System
2017	World Neurosurgery	Zolal	Probabilistic Tractography of the Cranial Nerves in Vestibular Schwannoma
2017	World Neurosurgery	Zhang	Preoperative Prediction of Location and Shape of Facial Nerve in Patients With Large Vestibular Schwannomas Using Diffusion Tensor Imaging‐Based Fiber Tracking
2019	World Neurosurgery	Churi	Correlation of Preoperative Cranial Nerve Diffusion Tensor Tractography With Intraoperative Findings in Surgery of Cerebellopontine Angle Tumors
2019	Neurology India	Samala	Effectiveness of Preoperative Facial Nerve Diffusion Tensor Imaging Tractography for Preservation of Facial Nerve Function in Surgery for Large Vestibular Schwannomas: Results of a Prospective Randomized Study
2020	American Journal of Neuroradiology	Castellaro	Multishell Diffusion MRI‐Based Tractography of the Facial Nerve in Vestibular Schwannoma
2020	Neurologia i Neurochirurgia Polska	Szmuda	Reliability of Diffusion Tensor Tractography of Facial Nerve in Cerebello‐Pontine Angle Tumours
2021	Journal of the Neurological Sciences	Ung	Accuracy and Outcomes of Diffusion Tensor Imaging Tractography in Resection for Vestibular Schwannoma for Facial Nerve Preservation
2021	Otology & Neurotology	Epprecht	Three‐Dimensional (3D) Printed Vestibular Schwannoma for Facial Nerve Tractography Validation
2022	World Neurosurgery	Pujol	Anticipating Facial Nerve Position Using Three‐Dimensional Tractography During the Preoperative Assessment of Cerebellopontine Angle Tumors
2023	Journal of Korean Neurosurgical Society	Zhang	Significance of Preoperative Nerve Reconstruction Using Diffusion Tensor Imaging Tractography for Facial Nerve Protection in Vestibular Schwannoma
2023	European Radiology	Shapey	Diffusion MRI of the Facial‐Vestibulocochlear Nerve Complex: A Prospective Clinical Validation Study
2024	Magnetic Resonance in Medicine	Hu	Preoperative Diffusion Tensor Imaging: Fiber‐Trajectory‐Distribution‐Based Tractography to Identify Facial Nerve in Vestibular Schwannoma

Abbreviations: FIESTA, fast imaging employing steady‐state acquisition; MRI, magnetic resonance imaging.

3.2. MR Acquisition Protocols

The extracted MR acquisition details are summarized in Table 2. Most studies (n = 19) used 3.0 T scanners, with Siemens (n = 10), Philips (n = 7), and GE (n = 4) being the primary vendors. Vendor information was not specified in one study. Coil configuration was poorly reported; 8‐channel and 32‐channel head coils were mentioned in 32% and 14% of studies, respectively.

TABLE 2.

Diffusion MR acquisition parameters: hardware, sequence, basic imaging parameters, and preprocessing.

Year	First author	Field strength (T)	Vendor	Head coil	dMRI sequence	TR/TE (ms)	Directions/b‐value (s/mm2)	Acceleration technique	Number of excitations	Scan time (min)	Structural imaging choice	dMRI preprocessing	Software
2006	Taoka	1.5	Siemens	NA	EPI	4900/85	NA/1000	NA	NA	NA	SSFP	No preprocessing	DT Visualizer v2
2011	Gerganov	3.0	Siemens	NA	NA	5600/86	12/1000	NA	1	NA	SSFP	No preprocessing	Brainlab iPlan v2.6
2014	Choi	1.5	Philips	6‐channel	EPI	10,398/72	32/1000	SENSE	1	NA	NA	Eddy current correction and motion correction by FSL	DTIStudio
2015	Yoshino	3.0	GE	8‐channel	EPI	17,000/65.6	30/1000	NA	1	NA	SSFP	Eddy current correction by proprietary software	DT Visualizer v2
2015	Yoshino	3.0	GE	8‐channel	EPI	17,000/65.6	30/1000	NA	1	NA	NA	Eddy current correction by proprietary software	DT Visualizer v2
2015	Wei	3.0	Siemens	8‐channel	NA	11,000/98	30/NA	NA	NA	NA	NA	No preprocessing	Brainlab iPlan v3.03
2016	Song	3.0	GE	8‐channel	NA	4000/82.2	30/1000	NA	2	9	NA	No preprocessing	3D Slicer
2016	Borkar	3.0	Philips	NA	EPI	4750–5600/80–90 a	15/800	SENSE	3	NA	NA	No preprocessing	DynaSuite Neuro v3.0
2016	Hilly	3.0	Philips	NA	NA	NA	32/NA	NA	NA	NA	NA	No preprocessing	FiberTrak (Philips) vR2.6.1
2017	Li	3.0	Siemens	8‐channel	EPI	4900/94	32/1000	NA	NA	NA	NA	No preprocessing	Medtronic StealthViz
2017	Zolal	3.0	Siemens	NA	RESOLVE	NA	20/800	NA	NA	12	NA	No preprocessing	FSL
2017	Zhang	3.0	GE	8‐channel	NA	6943/80.8	64/1000	NA	NA	NA	NA	No preprocessing	Brainlab iPlan v3.0
2019	Churi	3.0	Philips	NA	NA	2303/80	15/NA	SENSE	4	NA	T1	No preprocessing	FiberTrak (Philips)
2019	Samala	3.0	NA	32‐channel	NA	NA	NA/NA	NA	NA	NA	NA	No preprocessing	NA
2020	Castellaro	3.0	Philips	32‐channel	EPI	5408/98	12, 3, 64/300, 1000, 2000	NA	NA	23.4	NA	Denoising by MRTrix3; Topup and Eddy correction by FSL	MRTrix3
2020	Szmuda	1.5	Siemens	20‐channel	RESOLVE	4870/74,111	20/1000	NA	2	NA	T1+T2	No preprocessing	Medtronic StealthViz StealthDTI
2021	Ung	3.0	Siemens	NA	NA	NA	64/1000	NA	NA	NA	NA	No preprocessing	Brainlab iPlan Cranial v3.0 & 2.6
2021	Epprecht	3.0	Philips	8‐channel	EPI	NA	15/1000	NA	NA	3.1	NA	Eddy current correction by FSL	FSL and probtrackx
2022	Pujol	3.0	Philips	32‐channel	EPI	2614/80	32/800	NA	NA	6.4	NA	Reversed polarity‐based distortion correction	Brainlab iPlan
2023	Zhang	3.0	Siemens	NA	NA	11,700/96	30/1000	NA	NA	NA	NA	No preprocessing	Brainlab iPlan v3.05
2023	Shapey	3.0	Siemens	64‐channel	RESOLVE	1000/127	17, 30/700, 2000	NA	NA	NA	T2	Gibbs unringing, denoising, Topup/Eddy	MRTrix3
2024	Hu	3.0	Siemens	NA	EPI	6330−10,720/87−90	21 or 31/1000	NA	NA	NA	T1 + T2	Denoising, EPI distortion correction	MRTrix3 and unknown software for other algorithms

Abbreviations: dMRI, diffusion MRI; EPI, echo planar imaging; MR, magnetic resonance; NA, not available; RESOLVE, readout segmentation of long variable echo‐trains, a multishot sequence; s, second.

^a The original publication only stated these are minimum setups for TR/TE.

Diffusion sequences included single‐shot EPI (n = 9) and RESOLVE (n = 2), though 41% did not specify the sequence. In terms of b‐values, the majority (n = 13, 59%) utilized a single‐shell acquisition at 1000 s/mm². Gradient directions varied across studies, with most (n = 11, 50%) studies using 30–64 directions. Reported voxel sizes averaged 1.6 × 1.6 × 1.8 mm³ (Table 3). The reported TR ranged from 1000 to 17,000 ms, while TE ranged from 65.6 to 127.0 ms. TR and TE data were not reported in five studies. Acceleration techniques, number of averages, and number of excitations were rarely reported. Only five studies reported the dMRI scan time, with 23 min being the longest.

TABLE 3.

Field of view and imaging resolutions of diffusion imaging and structural imaging.

Year	Authors	dMR FOV (mm)	dMR resolution (mm)	Structural imaging FOV (cm)	Structural imaging resolution (mm)
2006	Taoka et al.	230× 230× 105	1.80 × 1.80× 3	20	0.39 × 0.39 × 1
2011	Gerganov et al.	NA× NA× 64	NA× NA× 1.6	22	0.49 × 0.49 × 0.5
2014	Choi et al.	240×240×175	2.5 × 2.5× 2.5	NA	NA
2015	Yoshino et al.	256× 256× NA	1.0 × 1.0× 2.5	20	0.39 × 0.39 × 0.8
2015	Yoshino et al.	256× 256× NA	1.0 × 1.0× 2.5	NA	NA
2015	Wei et al.	NA× NA× 120	NA× NA× 2	NA	0.6× 0.6 × 0.6
2016	Song et al.	200×200×48	1.56×1.56× 1.2	NA	Slice thickness = 0.8
2016	Borkar et al.	NA×NA×60	0.78 × 0.78 (reconstructed) × 1.5	NA	NA
2016	Hilly et al.	224×224×NA	2 × 2× 2	NA	NA
2017	Li et al.	230×230×NA	1.8 × 1.8× 1	NA	NA
2017	Zolal et al.	NA	2 × 2× 2	NA	NA
2017	Zhang et al.	224×224×120	NA× NA× 2	NA	Slice thickness = 0.8
2019	Churi et al.	195×195×60	1.75 × 1.75× 1.5	22	Slice thickness = 1
2019	Samala et al.	NA	NA	NA	NA
2020	Castellaro et al.	224×224×80	2 × 2× 2	NA	0.4× 0.4 × 0.6
2020	Szmuda et al.	230×230×NA	1.5 × 1.5× 1.5	MPRAGE: 26 × 26; TSE T2: 250 × 187	MPRAGE: 1.0 × 1.0 × 1.0; TSE T2: 0.6 × 0.6 × 2.0
2021	Ung et al.	256×256×NA	2 × 2× 2	NA	NA
2021	Epprecht et al.	320×200×NA	0.9 × 0.9× 1.5	NA	NA
2022	Pujol et al.	120×120×NA	2× 2× 2	NA	NA
2023	Zhang et al.	NA	NA× NA× 1	NA	NA
2023	Shapey et al.	NA	1.2 × 1.2× 1.2	Small FOV through IAC	0.5× 0.5 × 0.5
2024	Hu et al.	Group 1: 228×228×60; Group 2: 228×228×47	Group 1: 1.78×1.78× 1; Group 2: 1.5×1.5× 1.5	T1: 25 × 25; T2: 20 × 18	T1: 0.98 × 0.98 × 1; T2: 0.47 × 0.47 × 0.5

Abbreviation: NA, not available.

SSFP sequences were used in three studies for structural imaging. Another three studies reported the usage of T₁W and T₂W. Most of the studies did not specify which structural sequence they used (Table 2).

3.3. Preprocessing and Tractography

Tables 4 and 5 present the extracted preprocessing and tracking parameters. Raw dMR images were displayed in 36% of the studies. Most studies (n = 12) did not utilize or report imaging preprocessing. Motion/distortion and/or eddy current correction were reported in five studies, while denoising was performed in two studies. Brainlab iPlan was the software most frequently (n = 6) used, while a variety of other software options were reported.

TABLE 4.

Tracking algorithm overviews, and registration setups.

Year	First author	Raw dMRI image displayed	Signal modeling	Streamline generation algorithm	Deterministic or probabilistic?	ROI selection	ROI drawn on raw diffusion image?	Registration method	Fiber filtration method
2006	Taoka	Y	DTI	“Streamline”	Deterministic	IAC, CPA	NA	NA	Parameter thresholding
2011	Gerganov	N	DTI	Likely FACT	Deterministic	IAC, CN VII REZ	N	iPlan automatic fusion	Parameter thresholding + fiber pruning
2014	Choi	Y	DTI	FACT	Deterministic	IAC, CN VII REZ	N	NA	Parameter thresholding
2015	Yoshino	Y	DTI	“Streamline”	Deterministic	Porus acusticus, perpendicular to the assumed CN VII/VIII complex	NA	NA	Parameter thresholding
2015	Yoshino	Y	DTI	“Streamline”	Deterministic	Porus acusticus, perpendicular to the assumed CN VII/VIII complex	Y	NA	Parameter thresholding
2015	Wei	N	DTI	NA	Deterministic	“Tumor−brain surface,” IAC	NA	NA	Parameter thresholding + fiber pruning
2016	Song	N	DTI	NA	Deterministic	IAC, CN VII REZ	N	Rigid	Parameter thresholding
2016	Borkar	N	DTI	FACT	Deterministic	IAC, CN VII REZ	N	Affine	Parameter thresholding + fiber pruning
2016	Hilly	N	DTI	NA	Deterministic	NA	NA	NA	Parameter thresholding?
2017	Li	N	DTI	NA, proprietary software	Deterministic	IAC, CN VII REZ around the foramen of Luschka	N?	NA	Parameter thresholding?
2017	Zolal	N	Ball‐and‐stick	probtrackx	Probabilistic	IAC, CN VII REZ (symmetric to the healthy side)	N	Nonlinear symmetric image normalization method	Parameter thresholding
2017	Zhang	N	DTI	NA	Deterministic	“Tumor−brain surface,” IAC	N	Brainlab's built‐in “automatic fusion algorithm”	Parameter thresholding + fiber pruning
2019	Churi	Y	DTI	NA	Deterministic	From the CN VII REZ, multiple ROIs on cisternal CN VII course that can be visualized	N	Philips' built‐in automatic fusion algorithm	Parameter thresholding
2019	Samala	Y	NA	NA	NA	IAC, CN VII REZ	N	“Automatic fusion algorithm”	Manual pruning
2020	Castellaro	N	CSD	iFOD2, SD‐STREAM or SD‐STREAM‐L	Probabilistic for iFOD2, deterministic for SD‐STREAM and SD‐STREAM‐L	iFOD2: IAC, CN VII cisternal portion, CN VII REZ; SD‐STREAM: IAC, or IAC + CN VII REZ symmetric to the healthy side. Exclusion masks = tumor, temporal regions, medial and superior pons, CN V and “nonphysiologic structures”	N	Affine	Parameter thresholding + exclusion masks
2020	Szmuda	N	DTI	NA	Deterministic	IAC	N	NA	Parameter thresholding
2021	Ung	N	DTI	NA	NA	IAC, brainstem (location not specified)	N	NA	NA
2021	Epprecht	N	Ball and stick	probtrackx	Probabilistic	IAC, CN VII REZ	N	NA	Parameter thresholding
2022	Pujol	N	DTI	NA	Deterministic	IAC, pontomedullary sulcus, cochlear	N	NA	Parameter thresholding
2023	Zhang	N	DTI	NA	Likely deterministic	IAC, CN VII REZ	N	NA	Parameter thresholding
2023	Shapey	Y	CSD	iFOD2	Probabilistic	IAC, “brainstem border”	N	Affine	Parameter thresholding
2024	Hu	Y	CSD	iFOD1 + fiber‐trajectory (FTD)‐based tractography	Probabilistic	IAC, CPA using the unaffected side as reference	NA	NA	Parameter thresholding

Abbreviations: CN VII, facial nerve; CPA, cerebellopontine angle; CSD, constrained spherical deconvolution; dMRI, diffusion MRI; FACT, fiber assignment continuous tracking; FTD, fiber trajectory distribution; IAC, internal auditory canal; iFOD1, first‐order integration over fiber orientation distributions; iFOD2, second‐order integration over fiber orientation distributions; N, no; NA, not available; REZ, root entry zone; ROI, region of interest; SD_STREAM, streamlines tractography based on spherical deconvolution (SD); Y, yes.

TABLE 5.

Detailed tracking parameters and CN VII track inclusion/exclusion criteria.

Year	First author	FA threshold	Angular threshold (°)	Other thresholding parameters	Cisternal tracks manually pruned?	Cisternal tracks exclusion criteria	Brainstem fibers manually pruned?	Brainstem fiber pruning criteria/rationale	CN VII inclusion criteria
2006	Taoka	0.1	NA	NA	NA	NA	NA	NA	Tracks coursing between pons and the anterosuperior aspect of IAC
2011	Gerganov	0.1	NA	Track length threshold	Y	Not specified: tracks that disagree with “previous anatomy data”	Y	Longitudinal tracks or tracks that cross the midline	By the assumption of CN VII course
2014	Choi	0.15	70	/	NA	/	NA	/	Tracks connecting brainstem and IAC along the tumor capsule
2015	Yoshino	Variable, but not specified	30	Step length	NA	NA	NA	NA	Tracks connecting brainstem and IAC
2015	Yoshino	Variable, increased with an increment of 0.01	30	Step length	NA	NA	NA	NA	Tracks connecting brainstem and IAC, while coursing on the surface or inside of the tumor
2015	Wei	0.06−0.2	NA	Track length threshold	Y	Tracks that course to the temporal lobe, “traveled along the cerebellopontine cistern,” or have a “rigid curve”	Y	Longitudinal tracks or tracks that cross the midline	Tracts coursing between tumor‐brainstem border to cochlear, vestibule, or IAC, and conforms to the curvature shape of tumor. For the IAC part, the authors prefer to select tracks coursing on the anterior or anterosuperior wall
2016	Song	0.1	36.9	Track length threshold, initial tracking space, linear measurement threshold	Y	Tracks inside tumor, or perpendicular to the surface of the tumor, or tracks “outside the region”	Y	Longitudinal tracks such as “the pyramidal tract”	Tracts coursing between IAC and REZ, with shape “tangential” to the shape of tumor
2016	Borkar	0.12	60	/	NA	NA	Y	Longitudinal tracks or tracks that cross the midline	Tracks coursing between brainstem and IAC
2016	Hilly	0.1	90	Track length threshold	NA	NA	NA	NA	NA
2017	Li	Variable, increased with an increment of 0.02	NA	/	NA	NA	NA	NA	Tracks coursing between CN VII REZ and IAC
2017	Zolal	NA	NA	NA	NA	NA	NA	NA	By the assumption of CN VII course
2017	Zhang	0.06−0.20	NA	Track length threshold	Y	Tracks inside tumor	Y	Longitudinal tracks	Likely similar to Wei et al. 2015
2019	Churi	0.1	25−40	Track length threshold	NA	NA	NA	NA	By the assumption of CN VII course
2019	Samala	NA	NA	NA	NA	NA	Y	Longitudinal tracks or tracks that cross the midline	By the assumption of CN VII course
2020	Castellaro	0.1	30−45	Step size, number of streamlines, radius of spheric seeds	N	NA	NA	NA	By the assumption of CN VII course
2020	Szmuda	0.2, but might vary in cases	60, but might vary in cases	NA	NA	NA	NA	NA	Tracts connecting brainstem and IAC along the tumor capsule
2021	Ung	NA	NA	NA	NA	NA	NA	NA	NA
2021	Epprecht	NA	78	Number of samples, step length, number of steps	NA	NA	NA	NA	Tracts connecting brainstem and IAC
2022	Pujol	0.1	36.9	Track length threshold	NA	NA	NA	NA	By the assumption of CN VII course
2023	Zhang	0.1	NA	Track length threshold	NA	NA	NA	NA	Tracts connecting brainstem and IAC
2023	Shapey	NA	NA	Number of tracks, aborting threshold	NA	NA	NA	NA	Tracts connecting brainstem and IAC
2024	Hu	0.1 or 0.05 for some patients	30	In iFOD1: track length threshold, max length, seed number; in FTD: step size	N	NA	N	NA	Tracts connecting brainstem and IAC

Abbreviations: CN, cranial nerve; FTD, fiber trajectory distribution; IAC, internal auditory canal; iFOD1, first‐order integration over fiber orientation distributions; NA, not available; N, no; REZ, root entry zone; Y, yes.

In terms of signal models, DTI was the most commonly used method (n = 16, 73%). Since the year 2020, an increasing number of studies (n = 3) have utilized CSD. Deterministic algorithms were the mainstay (n = 17) approach for streamline generation, while there have been only four studies since 2017 using probabilistic methods. So far, only one study used geodesic tractography [8]. Unlike conventional local tracking methods that propagate streamlines based on stepwise voxel‐wise decisions, geodesic tractography computes globally optimal paths by considering the entire diffusion field simultaneously.

Manual ROI delineation and registration to structural images were common (n = 21), though only five studies explicitly described their registration methods. ROIs were typically placed at the IAC (n = 21) and REZ (n = 12), two of which used the healthy/unaffected side as a reference. Four studies opted for “tumor−brain surface.” FA thresholding (commonly 0.1) and angular thresholds (25°–90°) were used for fiber filtering. Other tracking parameters were also reported but varied, including track length, threshold step size, and number of streamlines.

Manual pruning of fiber tracts was reported in six studies. Of these, four specified exclusion criteria such as “tracks inconsistent with known anatomical data,” “tracks within the tumor,” “tracks traversing the CPA cistern,” or those displaying “rigid curvature.” Brainstem‐based pruning was also reported in six studies, typically eliminating “longitudinal tracks” or those “crossing the midline.” Inclusion criteria for CN VII tracks also varied. Nine studies defined CN VII as tracts connecting the brainstem to the IAC, with some refining this by requiring conformity to tumor morphology or the trajectory along the “tumor capsule.” Six studies inferred the CN VII course based on presumed anatomy but did not specify their criteria.

As shown in Table 6, visual intraoperative confirmation was used to validate CN VII locations in most studies (n = 21). Only one publication used neuroimaging/neuronavigation‐based confirmation. In one study [23], surgeons illustrated the course of CN VII by memory after surgery. In another work [24], surgeons manually drew the course of CN VII on a 3D‐printed tumor model after surgery. One study [25] overlaid the surgeon's drawing with tractography and quantified the maximal distance between the two. Only six of 22 studies specified the EMG threshold (ranging from 0.1 to 0.5 mA) used intraoperatively to confirm the CN VII course.

TABLE 6.

Recording and validation of the CN VII trajectory.

Year	First author	Recording of the preoperative CN VII location	Number of CN location subtypes	Multiple location recorded?	Validator/surgeon is blinded to tractography?	EMG thresholding (mA)	How to record the intraoperative CN VII course?	CN VII shape recorded?	Classification of CN shape
2006	Taoka	Variant of Sampath	6	NA	NA	NA	Surgeon's visual confirmation	N	NA
2011	Gerganov	Sampath	8	N	Y	1	Surgeon's visual confirmation	Y	Compact/flat
2014	Choi	Variant of Sampath	4	N	NA	NA	Surgeon's visual confirmation	N	NA
2015	Yoshino	Sampath	8	Y	NA	0.1−0.2	Surgeon's visual confirmation	N	NA
2015	Yoshino	Sampath	8	Y	NA	0.1−0.3	Surgeon's visual confirmation	N	NA
2015	Wei	Sampath	8	N	NA	0.1−0.5	Surgeon's visual confirmation	N	NA
2016	Song	Variant of Sampath	6	N	Y	0.1−1.0	Surgeon's visual confirmation	N	NA
2016	Borkar	Sampath	8	N	NA	NA	Surgeon's visual confirmation	N	NA
2016	Hilly	Variant of Sampath	3	N	Y	NA	Visual confirmation + illustration per memory by surgeon	Y	Good, effaced, or thin and redundant with splaying nerve fibers
2017	Li	Variant of Sampath	6	N	N	0.1−0.2	Neuronavigation‐based validation	N	NA
2017	Zolal	Variant of Sampath	12	Y	Y	NA	Surgeon's visual confirmation	N	NA
2017	Zhang	Variant of Sampath	6	N	NA	NA	Surgeon's visual confirmation	Y	Compact/flat (splayed)
2019	Churi	Variant of Sampath	8	Y	Y	NA	Surgeon's visual confirmation	Y	Compact/splayed/through tumor
2019	Samala	Sampath	8	N	N	NA	Surgeon's visual confirmation	Y	Compact/splayed
2020	Castellaro	Sampath	8	Y	NA	NA	Reviewer's visual confirmation on surgical video	N	NA
2020	Szmuda	Variant of Sampath	5	N	N	NA	Neuronavigation confirmation + surgical video review	Y	Compact/flat
2021	Ung	NA	NA	N	N	NA	Surgeon's visual confirmation	N	NA
2021	Epprecht	Variant of Sampath	4 (for sematic classification)	N	Y	NA	By memory, surgeons drew the course of CN VII on a 3D‐printed tumor model; a semantic classification was also used	N	NA
2022	Pujol	Variant of Sampath	7	N	NA	NA	Surgeon's visual confirmation	N	NA
2023	Zhang	Sampath	8	N	NA	0.1	Surgeon's visual confirmation	N	NA
2023	Shapey	Likely Sampath	3	N	Y	NA	Surgeon's visual confirmation	N	NA
2024	Hu	Sampath	8	N	Y	NA	Surgeon's visual confirmation	N	NA

Abbreviations: CN, cranial nerve; EMG, electromyography; FTD, fiber trajectory distribution; NA, not available; N, no; REZ, root entry zone electromyography; Y, yes.

Most (n = 21) of these observations were ultimately translated into the Sampath classification or its variant for further semantic comparison, while one study [24] did not specify. Nine studies used the original eight‐subtype Sampath classification system (e.g., antero‐superior, antero‐middle, postero‐inferior). Some studies [26] used mixed terms, such as “anterior,” “antero‐supero‐middle,” and “antero‐middle.” For consistency, we interpreted “anterior” as “antero‐middle” in these cases. However, in studies with fewer than eight location subtypes [23, 27], we retained the original phrasing of “anterior,” as it was not clear if they combined the definition of “antero‐middle” and “antero‐inferior.” Only 27% (n = 6) of the studies reported multiple CN VII locations (e.g., both “antero‐middle” and “antero‐inferior”). One study [23] focused solely on the lateral third of the CN VII course.

The shape of CN VII was described in six studies, most commonly as “compact” or “flat,” often alongside “splayed” in cases of tumor distortion or effacement. Thirteen studies did not report EMG stimulation thresholds. For the five studies reporting thresholds, the range was 0.1−0.5 mA.

3.4. Demographics and Tracking Outcomes

The sample size and statistical findings for each study are given in Tables 7 and 8. Sample sizes ranged from 5 to 47 (mean = 20.2). Koos Grade III–IV tumors were present in 55% of cases, among which the percentage of Koos III/IV enrollment ranged from 40% to 100%. Seven studies exclusively included Koos III/IV tumors, with CN VII success rate ranging from 81% to 97%. While most studies (n = 18, 82%) provided radiological measurements of tumor size, the methods used to define tumor size were inconsistent. Therefore, we recorded the maximal value of reported tumor dimensions as a surrogate for tumor size across studies. The resulting diameters ranged from 1.68 to 4.26 cm (mean = 3.1 cm).

TABLE 7.

Demographics and basic CN VII validation measurements.

Year	First author	Country	N of patients	Koos Grade III−IV rate	Mean tumor diameter (cm)	Percentage of cases that successfully identified the CN VII location intraoperatively	Percentage of cases with successful preoperative CN VII rendering	Percentage of cases with CN VII location prediction corresponding with intraoperative findings	True positive count for CN VII	False positive count for CN VII	False negative count for CN VII
2006	Taoka	Japan	8	NA	2.96 ± 1.08	7/8 (0.88)	7/8 (0.88)	5/6 (0.83)	5	1	2
2011	Gerganov	Germany	22	100%	2.70 ± 0.78	1	1	20/22 (0.91)	20	2	2
2014	Choi	South Korea	11	64%	2.47 ± 0.73	1	1	1	11	0	0
2015	Yoshino	Japan	21	NA	2.98 ± 1.10	1	17/21 (0.81)	3/21 (0.14)	3	17	18
2015	Yoshino	Japan	11	NA	2.98 ± 0.68	1	10/11 (0.91)	3/10 (0.3)	3	9	8
2015	Wei	China	23	100%	3.24 ± 0.98	22/23 (0.96)	22/23 (0.96)	21/22 (0.95)	21	0	0
2016	Song	China	15	NA	3.19 ± 0.85	1	14/15 (0.93)	13/14 (0.93)	13	1	1
2016	Borkar	India	20	NA	4.26 ± 0.78	1	19/20 (0.95)	16/18 (0.89)	16	2	2
2016	Hilly	Canada	21	NA	3.17 ± 0.71	1	20/21 (0.95)	19/20 (0.95)	18	1	1
2017	Li	China	19	100%	4.13 ± 0.76	1	18/19 (0.94)	16/18 (0.89)	17	1	1
2017	Zolal	Germany	21	100%	NA	1	1	17/21 (0.81)	18	17	4
2017	Zhang	China	30	100%	3.40 ± 0.85	29/30 (0.97)	1	29/30 (0.97)	29	0	0
2019	Churi	India	31	NA	3.80 ± 0.95	1	1	26/31 (0.84)	28	5	6
2019	Samala	India	47	NA	>3	39/40 (0.98)	1	39/39 (1.00)	39	0	0
2020	Castellaro	Italy	5	40%	1.68 ± 0.26	1	0.2−1	1/5 (0.2)–4/5 (0.8)	2−10	NA	NA
2020	Szmuda	Poland	38	53%	2.88 ± 1.10	1	34/38 (0.89)	31/34 (0.91)	31	3	3
2021	Ung	USA	11	NA	2.82	1	1	10/11 (0.91)	NA	NA	NA
2021	Epprecht	USA	16	53%	1.84 ± 0.6	19/20 (0.95)	16/20 (0.8)	15/16 (0.94)	15	1	4
2022	Pujol	Spain	10	NA	NA	1	1	NA	NA	NA	NA
2023	Zhang	China	27	100%	3.54 ± 0.87	1	27/30 (0.9)	25/27 (0.93)	25	2	2
2023	Shapey	UK	5	100%	NA	1	NA	NA	NA	NA	NA
2024	Hu	China	33	88%	3.30 ± 1.17	1	1	Group 1: 13/14 (0.93); Group 2: 13/19 (0.68)	Group 1: 13; Group 2: 13	Group 1: 1; Group 2: 6	Group 1: 1; Group 2: 6

Note: All the data represent mean±SD unless otherwise indicated.

Abbreviations: CN, cranial nerve; NA, not available; N, no.

TABLE 8.

Inferred CN VII validation outcomes.

Year	First author	CN VII location sensitivity (TP/[TP+FN])	CN VII location positive predictive value (TP/[TP+FP])	CN VII false discovery rate (FP/[TP+FP])	Percentage of cases with CN VII shape prediction corresponding with intraoperative findings	Percentage of cases with successful preoperative CN VIII rendering	Percentage of cases with CN VIII tractography location corresponding with intraoperative findings	Percentage of cases with CN VIII tractography shape corresponding with intraoperative findings
2006	Taoka	0.71	0.83	0.17	NA	NA	NA	NA
2011	Gerganov	0.91	0.91	0.09	NA	NA	NA	NA
2014	Choi	1	1	0	0	NA	0	NA
2015	Yoshino	0.14	0.15	0.85	NA	17/21 (0.81)	13/21 (0.62)	NA
2015	Yoshino	0.27	0.25	0.75	NA	10/11 (0.91)	6/10 (0.6)	NA
2015	Wei	1	1	0	0	NA	4/4 (1.00)	4/4 (1.0)
2016	Song	0.93	0.93	0.07	NA	0	NA	NA
2016	Borkar	0.89	0.89	0.11	NA	NA	NA	NA
2016	Hilly	0.95	0.95	0.05	NA	0	NA	NA
2017	Li	0.94	0.96	0.05	NA	0	0	NA
2017	Zolal	0.82	0.51	0.49	NA	18/21 (0.86)	6/18 (0.33)	NA
2017	Zhang	1	1	0	NA	NA	0	NA
2019	Churi	0.82	0.84	0.15	24/31 (0.77)	15/31 (0.48)	4/15 (0.27)	5/15 (0.33)
2019	Samala	1	1	0	NA	NA	NA	NA
2020	Castellaro	NA	NA	NA	NA	0	NA	NA
2020	Szmuda	0.91	0.91	0.09	NA	0	NA	NA
2021	Ung	NA	NA	NA	NA	0	NA	NA
2021	Epprecht	0.79	0.94	0.06	NA	0	NA	NA
2022	Pujol	NA	NA	NA	NA	0	NA	NA
2023	Zhang	0.93	0.93	0.07	NA	0	NA	NA
2023	Shapey	NA	NA	NA	NA	0	NA	NA
2024	Hu	Group 1: 0.93; Group 2: 0.68	Group 1: 0.93; Group 2: 0.68	Group 1: 0.07; Group 2: 0.32	NA	0	NA	NA

Abbreviations: CN, cranial nerve; FN, false negative; FP, false positive; NA, not available; TP, true positive.

Tractography was rendered successfully in most studies, though their concordance with intraoperative findings averaged 83% (SD 22%) across studies. Notably, two studies [28, 29] by the same group reported “success rates” as low as 14–30%. The mean sensitivity, PPV, and FDR of CN VII tractography were 0.82 ± 0.24, 0.82 ± 0.25, and 0.18 ± 0.25, respectively. Most studies (n = 21) were unable to distinguish CN VII shape on tractography. In addition, only four studies claimed to render CN VIII on tractography, with only two comparing pre‐ versus intraoperative findings.

3.5. Bias Analysis

A curated QUADAS‐2 form was made specific to the review question (Table 9). We aimed to analyze and quantify the bias of the included studies. After careful evaluation (Table 10), we found a strikingly high bias across all four domains (86% in patient selection, 91% in index test, 100% in reference standard, and flow and timing). Applicability concerns existed in 68% of the studies in the patient selection: 68% domain and 100% in the reference standard.

TABLE 9.

QUADAS‐2 form for the current study.

Review question: Can facial nerve tractography accurately map the facial nerve course in preoperative Koos Grade III–IV vestibular schwannoma where conventional MR (T1, T2, or SSFP) cannot?
Domain	Patient selection	Index test	Reference standard	Flow and timing
Description	Included patients should be preoperative Koos Grade III or IV VS who are treatment‐naive; sporadic VS only; NF‐2 excluded	The study should perform facial nerve tractography, which is translated into the Sampath classification system	Ideally, intraoperative electrophysiology confirmation with the guidance of neuronavigation, which is recorded by the Sampath classification system (eight subtypes, i.e., antero‐superior, antero‐middle, antero‐inferior, postero‐superior, postero‐middle, postero‐inferior, superior, inferior), should be the “gold standard.”
Signaling questions	Signaling question 1: Was a consecutive or random sample of patients enrolled? Was a consecutive or random sample of sporadic non‐NF‐2 VS patients enrolled over a consecutive calendar period (≥ 6 months) or a prospectively defined registry sampling frame? Signaling question 2: Was a case–control design avoided? Did the report of CN VII tractography avoid healthy controls or non‐VS patients? Signaling question 3: Did the study avoid inappropriate exclusions? Did the report of CN VII tractography success rate exclude failed cases or large VS?	Signaling question 1: Were index test results interpreted without knowledge of the results of the reference standard? Were tractography results recorded without prior knowledge of the intraoperative results? Or were tractography performed on a post hoc basis when the intraoperative results were obtained? Signaling question 2: If a threshold was used, was it prespecified? Were tracking parameters (b‐values, number of directions, ROI, FA threshold, step size, etc.) prespecified in a clear manner before tractography was performed?	Signaling question 1: Is the reference standard likely to correctly classify the target condition? Does certain electrophysiology threshold setups correctly classify the facial nerve? Is it possible for surgeons to correctly locate the facial nerve during surgery? By visual confirmation with the neuronavigation, is it possible to correctly classify the facial nerve courses (e.g., not misclassifying antero‐superior into antero‐middle)? Signaling question 2: Were the reference standard results interpreted without knowledge of the results of the index test? Did the study explicitly state that the intraoperative CN VII course was recorded without knowledge of the preoperative tractography results?	Signaling question 1: Was there an appropriate interval between the index test and reference standard? Was the facial nerve tractography performed within 3 months preoperatively? Signaling question 2: Did all patients receive the same reference standard? Were all patients’ facial nerve course recorded using intraoperative electotrophysiology ± neuronavigation ± Sampath classification?
Concerns about applicability	Applicability: Are there concerns that the included patients and setting do not match the review question? Low concern if ≥ 95% patients are Koos III−IV.		Are there concerns that the target condition as defined by the reference standard does not match the condition? Is the full CPA cistern course of CN VII evaluated for Sampath classification system? Does Sampath classification fully depict the course of CN VII?

Abbreviations: CN, cranial nerve; CPA, cerebellopontine angle; FA, fractional anisotropy; MR, magnetic resonance; NF, neurofibromatosis; QUADAS, Quality Assessment of Diagnostic Accuracy Studies; ROI, region of interest; SSFP, steady‐state free precession; VS, vestibular schwannoma.

TABLE 10.

QUADAS‐2 bias analysis of included studies.

	Risk of bias				Applicability concerns
Study	Patient selection	Index test	Reference standard	Flow and timing	Patient selection	Index test	Reference standard
1	unclear	unclear	unclear	unclear	unclear	NA	unclear
2	unclear	unclear	unclear	unclear	low		unclear
3	unclear	unclear	unclear	unclear	high		unclear
4	high	low	unclear	unclear	unclear		unclear
5	unclear	low	unclear	unclear	unclear		unclear
6	unclear	unclear	unclear	unclear	low		unclear
7	unclear	unclear	unclear	unclear	unclear		unclear
8	high	unclear	unclear	high	unclear		unclear
9	unclear	high	unclear	high	unclear		unclear
10	unclear	unclear	unclear	unclear	low		unclear
11	unclear	unclear	unclear	unclear	low		unclear
12	unclear	unclear	unclear	high	low		unclear
13	unclear	unclear	unclear	unclear	unclear		unclear
14	high	unclear	high	high	unclear		unclear
15	unclear	unclear	unclear	unclear	high		unclear
16	unclear	unclear	high	unclear	high		unclear
17	high	high	high	unclear	unclear		unclear
18	unclear	unclear	unclear	unclear	unclear		unclear
19	low	unclear	unclear	unclear	unclear		unclear
20	low	unclear	unclear	unclear	low		unclear
21	low	unclear	high	high	low		unclear
22	high	unclear	unclear	unclear	high		unclear

Abbreviations: NA, not available; QUADAS, Quality Assessment of Diagnostic Accuracy Studies.

4. Discussion

This systematic review highlights the substantial variabilities in MR acquisition, preprocessing, tractography, validation methods, and outcome report standards in preoperative CN VII tractography for VS. Since 2016, six systematic reviews [10, 12, 13, 14, 15] (Table 11) have investigated the reliability and application of CN VII tractography in the preoperative evaluations of VS. While reported “success rates” in the literature often exceed 87% [10, 12, 13, 14, 15], our analysis revealed that these figures might overestimate the true reliability of CN VII tractography in VS, particularly in larger tumors. In practice, preoperative CN VII tractography is most valuable in larger VS, as the nerve is usually identifiable on structural scans in small‐to‐medium sized tumors, obviating the need of tractography [30]. Larger tumors also present greater surgical complexity, making precise CN VII mapping most critical in these cases. As such, the “true” value of preoperative CN VII tractography may be overestimated if “success rates” are inflated by data pooled from smaller, less challenging tumors. Below, we summarize our key observations from the extracted data to clarify the current status of CN VII tractography in VS.

TABLE 11.

Existing systematic reviews on CN VII tractography in VS.

Year	First author	Number of included studies	Pooled analysis? (Y/N)	Pooled sample size	Pooled tractography concordance rate	Noticed tractography failure? (Y/N)
2016	Ung	3	Y	36	32/36 (89%)	Y
2018	Savardekar	14	Y	234	205/234 (87%)	Y
2019	Baro	17	Y	214	187/214 (87%)	Y
2021	Wu	16	Y	330	89%	Y
2023	Bubeníková	8	Y	149	89%	Y

Abbreviations: CN, cranial nerve; N, no; VS, vestibular schwannoma; Y, yes.

4.1. MR Acquisition Protocols

Substantial methodological heterogeneity across studies limits direct comparison of acquisition parameters and precludes identification of a universally optimal TR/TE combination. From a signal perspective, diffusion‐weighted acquisitions inherently incur substantial signal attenuation; therefore, minimizing TE for a given b‐value and gradient hardware constraints is generally desirable to preserve SNR. In addition, longitudinal T₁ recovery must be considered. For a given tissue T₁, SNR efficiency is theoretically maximized when TR is approximately 1.26 × T₁ [31], reflecting a balance between signal recovery and acquisition time. For example, assuming a representative cranial nerve T₁ of ∼1000 ms at 3T, this relationship suggests an optimal TR of approximately 1.2–1.3 s, although practical diffusion protocols typically employ longer TRs to accommodate slice coverage requirements.

Reported in‐plane resolutions span 0.78 × 0.78 mm to 2 × 2 mm, with only two studies employing a voxel just under 1 × 1 mm. Yet, the cisternal segment of CN VII measures approximately 1 mm in diameter and courses immediately adjacent to CN VIII, reaching its widest separation (1–2 mm) at the pontomedullary sulcus [32]. Considering this, such in‐plane resolutions may be insufficient to distinguish CN VII from CN VIII, even on a healthy person, due to partial volume effects. Partial volume effect likely worsens in the presence of tumors, where the proximity of nerves, CSF, and tumor tissues combine to produce an averaged diffusion signal within a voxel. This problem worsens in large VS, especially when CN VII is “flattened” or “splayed,” further diluting the contribution of nerves to the overall signal and increasing tractography errors. Higher in‐plane resolution could mitigate these effects, but at the cost of longer scan times and lower SNR.

Advanced multishell dMRI combined with multitissue signal models can also, in theory, disentangle the contributions of CSF, nerve, and tumors, thereby reducing partial volume effects [33]. However, only two publications employed multishell acquisitions, due partly to the recent emergence of this technique. In addition, acquiring dMRI data at multiple TEs may provide further improvements in our ability to isolate signals from cranial nerves from surrounding structures [21]; however, this approach has yet to be investigated in this context. Finally, employing larger b‐values (≥2000 s/mm²) may improve our ability to accurately capture complex white matter fiber orientation information [8]; however, this comes at the expense of SNR in the raw images. Only two studies employed b‐values greater than 1000 s/mm² in combination with lower b‐values (e.g., 300 and 1000 s/mm²), enabling simultaneous sampling of anisotropic (white matter‐like) and isotropic (CSF‐like) signal components. Both studies implemented multitissue CSD modeling using MRtrix.

The number and distribution of diffusion‐encoding directions is another critical acquisition variable. It has been shown that 30 directions should be sufficient for relatively accurate tractography results in tensor‐based tractography, but the ultimate benefit of very high angular sampling remains an open question [20, 34, 35]. Hence, no universal standard exists for diffusion directions, especially in the setting of VS.

Increasing diffusion gradient angular sampling and in‐plane resolution inevitably prolong acquisition time, increasing susceptibility to motion and limiting clinical feasibility. Across studies, reported diffusion scan times ranged from 6.4 to 23.4 min, while comprehensive preoperative protocols that integrate multiple contrasts may extend substantially longer in clinical practice. These practical constraints underscore the need to balance spatial resolution, SNR, and acquisition time to maintain image quality without overburdening patients.

Variations across hardware (vendor, field strength, coils), pulse sequences (single‐ vs. multishot), and image reconstruction methods (e.g., acceleration approaches) may introduce significant bias across studies, which are difficult to quantify. When possible, acquisition protocols should be standardized. In addition, statistical “harmonization” approaches [19] have shown promise for large white matter tracts, but have yet to be evaluated in the context of cranial nerve dMRI.

4.2. Preprocessing and Tractography

Only a handful of studies present the raw diffusion‐weighted images, a key element for methodological transparency. In practice, the performance of dMRI CN VII tractography is limited by EPI‐related geometric distortions, subject motion, and noise [8, 36]. For this reason, an increasing number of groups have incorporated preprocessing in their workflow, but this procedure has yet to be standardized [8, 36]. In our review, very few papers reported techniques such as motion correction, eddy‐current correction or denoising, and none formally quantified whether these steps improved SNR, contrast‐to‐noise ratio (CNR), or tractography accuracy in VS. It is noteworthy that incomplete or absent distortion correction among other preprocessing steps can likely contribute to both false negatives (missed CN VII) and false positives (mislocalized streamlines at the brainstem or IAC). Establishing and validating a consensus preprocessing pipeline should, therefore, be a priority for future work.

After preprocessing, within‐voxel diffusion modeling (e.g., using DTI or CSD) is performed. However, this process may be prone to bias, particularly in the presence of partial volume effects in larger VS. Once orientation information is estimated, deterministic algorithms—used in most studies in this review—trace peak orientations from voxel to voxel to generate streamlines. In contrast, probabilistic algorithms, though less commonly used, account for the uncertainty inherent in estimating these orientations. In either case, these stepwise tracking methods can compound the errors of signal modeling as they propagate. Unlike conventional stepwise methods, geodesic and global tractography incorporate more global information from the diffusion signal. Geodesic tractography [37] identifies pathways by computing shortest paths through a diffusion‐defined manifold [8]. Although not included in these studies, global tractography [8] reconstructs all fiber tracts simultaneously by fitting a model to the entire dataset, promoting consistency that may improve performance in CN VII tractography. CSD probabilistic tracking has gained traction in recent years, implemented in open‐source research software (MRTrix [38, 39]) as well as proprietary clinical workstations (Medtronic Stealthstation and Brainlab). However, CSD probabilistic tracking has yet to be established as a “gold standard” despite its theoretical advantages. Accordingly, future studies should include head‐to‐head comparisons of CSD probabilistic and conventional DTI deterministic tracking within the same VS cohorts to rigorously evaluate relative performance.

Independent of the signal model or tracking algorithm, ROIs serve as the starting and stopping points for fiber tracking and impose anatomical priors that constrain or filter streamline generation within specific brain regions. Due to the low‐resolution nature of diffusion data, it is challenging to delineate ROIs on diffusion scans. Therefore, diffusion data are usually registered to high‐resolution structural images. There are several nuances in this process. First, the selection of ROI is critical. Most of the studies used IAC and CN VII REZ as ROIs, but few studies specified the details of ROI drawing (e.g., full IAC coverage vs. selected slices, inclusion of the cochlear, and the boundary of the tumor−brain interface), raising questions of reproducibility. Third, few studies have validated the registration accuracy. The discrepancy of contrast and imaging quality (e.g., distortions) between two maps can contribute to misalignment, producing false negatives or false positives. As such, drawing directly on diffusion images might be the most precise approach.

Even with carefully optimized pipelines, spurious fibers are common and often require postprocessing or pruning. As such, some studies used manual pruning to remove anatomically implausible or undesired tracts. While we agree that streamlines extending into the temporal lobe or forming rigid, anatomically implausible curves likely represent false positives, other criteria used for pruning warrant closer scrutiny. For example, Song et al. [40] with a 93% success tracking rate filtered out tracks coursing through the tumor, by which standard true positives of CN VII might be eliminated, because tumor invasion or encapsulation of CN VII is possible [7]. Other studies excluded “longitudinal tracts” or streamlines crossing the midline within the brainstem, assuming these represented corticospinal tracts, the medial lemniscus, or transverse pontine fibers. However, we suspect that some midline‐crossing streamlines could in fact be transverse pontine fibers, while “longitudinal tracts” may represent the corticobulbar tract, spinothalamic tract, or medial lemniscus [41, 42].

Of note, the intrabrainstem CN VII course features a slender path (approximately 1 mm diameter) and two roughly 90‐degree turning angles within an area about 1×1 mm² [42]. Due to partial volume effects, it is challenging to capture such intricate anatomy when the most common scanner resolution of clinical MR systems is within the range of 1–2 mm. Therefore, the tracking algorithm might mistakenly connect intrabrainstem portion of the CN VII with other adjacent white matter tracts. By deleting these tracks, we might risk removing the “true positives.” Regardless, whether using the intrabrainstem tracks to identify CN VII in the presence of VS should be investigated more extensively in future research.

Constant current stimulation is widely used in clinical practice, where CN VII can be mapped intraoperatively using compound muscle action potential monitoring. However, intraoperative stimulation thresholds were largely under‐reported across the included studies. Existing best‐practice recommendations suggest that low stimulation thresholds (e.g., ≤0.1 mA) are most informative for confirming functional proximity/integrity of CN VII [43, 44]. Among the studies that reported thresholds, values ranged from 0.1 to 0.5 mA, with all reported thresholds at or above 0.1 mA. Notably, none of the included studies performed systematic point‐to‐point mapping of the full CN VII course using stimulation. Instead, the Sampath classification (Table 6) was the most commonly used system to summarize CN VII position relative to the tumor for both tractography and intraoperative observations.

Although this system offers a straightforward approach, several important limitations warrant consideration. First, six studies noted that CN VII can assume multiple trajectories from medial to lateral. Interestingly, Hilly et al. specifically evaluated only the lateral 1/3 of the CN VII when applying the Sampath classification [23]. This raises the question of whether other authors—who provide only a single course label—adequately captured the nerve's entire path. Second, most studies relied on intraoperative visual confirmation, a method prone to inaccuracies, as the relative superior/inferior CN VII position to the tumor can appear unclear after tumor debulking. To mitigate this, Li et al. attempted to employ intraoperative neuronavigation to confirm the course of CN VII [45]. Moving forward, we recommend neuroimaging‐based validation alongside multicourse Sampath reporting to better capture CN VII's true complexity. However, it should be noted that borderline scenarios (e.g., the facial nerve on the border of antero‐middle and anterior‐superior) might still bring challenges to reporting.

4.3. Demographics and Tracking Outcomes

Several methodological concerns also arise when reviewing reported tracking outcomes. First, as noted above, pooling data from different studies can inflate “success rates,” especially if many participants have small‐to‐medium tumors that are easier to track. Second, most studies used a single “concordance rate” or “success rate” as its reliability benchmark. Yet, CN VII tractography can produce both false positives (or “spurious fibers”) or false negatives (“unseen fibers”), neither of which is systematically reported in these studies. Such oversights can greatly inflate sensitivity, PPV, and deflate FDR. As an illustration of this variability, two studies by the same group (Yoshino et al., 2015) [28, 29] reported comparatively low success rates. However, direct comparison of these studies with higher‐performing series is limited by incomplete reporting of acquisition parameters, preprocessing steps, and validation criteria, issues that are discussed further in the bias analysis section. Third, some studies reported observing the shape of CN VII on preoperative tractography. In our view, however, CN VII tractography mainly provides the coordinates of possible fiber pathways, making it unlikely to yield detailed morphological information such as shape. It is still unclear if we can infer CN VII shape from diffusion biomarkers (e.g., FA). Finally, a few groups report simultaneous reconstruction of CN VIII and CN VII, yet partial volume effects from the proximity of these two nerves make reliable differentiation extremely challenging, a limitation acknowledged across the literature.

4.4. Bias Analysis

The QUADAS‐2 bias analysis supported our extracted results. As for patient selection, we believe that CN VII tractography has the most value for Koos Grade III/IV patients. CN VII tractography provides little or no value in Koos Grade I/II patients, as conventional structural imaging can map the CN VII course in smaller tumors. In addition, CN VII is less likely to be “splayed out” in Koos Grade I/II patients. By excluding “difficult” cases, or enrolling Koos Grade I/II or nonconsecutive patient cohorts, studies may inadvertently inflate the accuracy of CN VII tractography. Regarding the index test (methodology of CN VII tractography), an ideal study should explicitly state the MR acquisition parameters and tracking parameters, as well as the blinding of the reference standard (i.e., intraoperative findings). In terms of reference standard, flow, and timing, it is important that CN VII results are blindly validated by intraoperative neuronavigation (instead of visual confirmation), along with a clear protocol of electrophysiology stimulation threshold. Sampath classification system offers a shortcut for validation, but it is extremely limited by its oversimplifying nature. Of the seven studies that exclusively included Koos Grade III/IV tumors (low patient applicability concern), only two studies [46, 47] had low bias in patient selection. However, their bias for other domains remained either unclear or high. Therefore, none of the studies had low bias across all domains. The true accuracy of CN VII tractography in VS remains unclear, as the quality of evidence remains low.

4.5. Actionable Recommendations

Based on the available literature and established diffusion signal principles, a reasonable diffusion MRI protocol for CN VII in VS would prioritize high in‐plane spatial resolution (ideally approaching submillimeter where feasible), multishell acquisition when clinically practical, adequate angular sampling, and robust preprocessing, including distortion and eddy‐current correction with careful registration to high‐resolution structural images. Incomplete distortion correction or imprecise registration may directly contribute to both false negatives (missed CN VII) and false positives (mislocalized streamlines at the brainstem or IAC). Advanced signal modeling approaches such as CSD and probabilistic streamline generation are promising, but current evidence remains insufficient to establish them as a gold standard. Prospective head‐to‐head comparisons within comparable VS cohorts are needed to define optimal acquisition and modeling strategies.

4.6. Limitations

This review is not without limitations. First, we searched for a single database (PubMed), which may have led us to miss relevant studies indexed in other databases (e.g., Embase, Scopus, Web of Science). Second, our search prompt was relatively narrow, potentially excluding works that examined CN VII tractography in VS but did not explicitly mention those keywords in their title or abstract. Third, we did not conduct citation tracking of the references in the papers included, which is a common strategy for identifying additional studies. Expanding the search to multiple databases and performing citation tracking would provide more comprehensive coverage. However, since CN VII tractography remains a specialized technique that is not widely adopted in clinical practice, the likelihood of missing major studies should be low. Nevertheless, our approach was still consistent with the PRISMA guideline, given its transparency and limitations discussed above. In addition, QUADAS‐2, built for diagnostic studies, has limited applicability in this unique topic as the true reference standard in many included studies were lacking, yielding many assessments as “unclear.”

5. Conclusion

While diffusion MR tractography holds considerable promise for localizing CN VII in VS, numerous technical and methodological variables complicate the validity of current “success rate” estimates. Recognizing and measuring false positives and false negatives may represent key steps in refining and accurately assessing the reliability and repeatability of CN VII tractography in VS. In the meantime, ongoing research should be focusing on: (1) improving the MR acquisition to improve data quality; (2) systematically evaluating the impact of preprocessing steps and signal modeling on the fiber tracking results; (3) developing novel tracking algorithms tailored for this challenging application (e.g., that incorporate a priori knowledge of the anatomy and/or information from other MRI contrast mechanisms); and (4) developing more comprehensive validation methods in the setting of the operating room.

Funding

This study was funded by the Barrow Neurological Foundation.

Conflicts of Interest

The authors declare no conflicts of interest.