Mapping and anatomo-surgical techniques for language preservation in brain tumours (Review)

Abstract

Glioma resection surgeries require balance between language functioning protection and maximal tumour removal. Surgical planning is key for the success of resection procedures and the techniques involved are often associated with cortical mapping and tissue differentiation. These techniques include direct electrical stimulation, stereo-electroencephalography, functional magnetic resonance imaging, diffusion tensor imaging, functional ultrasound, hyperspectral imaging and intraoperative optical imaging. The present review explores the anatomical features of language-eloquent areas and the techniques and technologies developed to map cortical areas and distinguish tumour tissue from functional tissue. The benefits and limitations of each technique are discussed, along with their intraoperative and preoperative uses. All techniques are discussed in the context of their effectiveness in the preservation of cortical areas responsible for language processing.

1. Introduction

Language is particularly important in human cognition, communication and thinking, with the appearance of written language having been an important advancement in recorded history (1,2). Extensive research has been conducted to explore the association between language functioning and associated cortical regions. The key cortical regions involved in language functioning have also been implicated in other cognitively demanding tasks such as numerical processing tasks (3-5).

The safest and most common surgical practice for the treatment of intracranial malignant tumours is the maximal safe resection of the tumour. The removal of brain tumours has been shown to improve patient health and survival outlook, along with a decrease in intracranial pressure and greater dependence from associated treatment (3,6,7). It is important for neurosurgeons to try and preserve as much of language functioning as possible during the resection of brain tumours located in or near eloquent language areas. The method used for preoperative mapping (Wada tests) have previously lacked the precision and reliability needed to accurately predict postoperative results. Brain tumours, such as gliomas, exhibit marked heterogeneity and cortical organization varies notably across individuals. In some cases, language function may remain intact even when the tumour is located within eloquent language areas, making it difficult to simultaneously balance maximal tumour resection with preservation of language function (6,8-10).

Awake craniotomy with direct electrical stimulation (DES) is one of a number of methods previously developed to improve brain mapping ability and accurately locate language-eloquent areas during surgical operations. Despite allowing surgeons to observe how language function is being affected during surgery and further improving overall surgical precision, difficulties remain regarding the treatment of multilingual patients, as each language may be localised in a different language area. Due to these difficulties, surgeons need to take preoperative and intraoperative precautions, such as diffusion tensor imaging (DTI) and stereo-electroencephalography (SEEG) discussed below (3,4,5,2,8,11-19).

The mapping of language function in multilingual patients is particularly difficult, especially in individuals with lower language proficiencies or languages learnt later in life. Linguistic representation in these languages may develop cortical networks that are more diffusely distributed compared with those of native or dominant languages (5). In those cases, the non-native languages acquired are more likely to utilise bilateral or right-hemispheric structures that results in a lack of overlapping activation patterns.

Due to this, intraoperative mapping for multilingual patients should aim to incorporate language-specific tasks and involve consultation with bilingual neuropsychologists. If language-specific procedures are not followed, there may be an under-identification of eloquent cortical areas, which increases the risk of postoperative language regression (20). The present review provides an overview of current and developing technologies dedicated to the protection of language function areas during brain tumour surgery.

2. Anatomy of language areas in the brain

Language-processing regions within the brain encompass a complex and interconnected network of cortical and subcortical regions. Historically, this network has been described as comprising two key language centres, the Broca's and Wernicke's areas, but it also includes numerous additional cortical regions and subcortical tracts.

Traditional language regions

Broca's and Wernicke's areas are traditionally considered to be key language-processing regions in the brain and are responsible for language expression and comprehension, respectively (21-23). Located in the posterior portion of the inferior frontal gyrus (IFG), the Broca's area is responsible for language expression, while Wernicke's area located in the superior temporal gyrus exhibits the function of language comprehension and the decoding of spoken language for its semantic meaning (21-26). However, contemporary research has begun to alter our understanding of the functions and cortical regions associated with these areas. In addition to being considered a suitable reference for the entirety of the IFG, Broca's area is now hypothesised to serve roles in both phonological and semantic processing, although its dominant function remains subject to debate. The anatomical regions included in Wernicke's area now also include angular gyrus, superior marginal gyrus, middle temporal gyrus and inferior temporal gyrus. Furthermore, the Broca's and Wernicke's areas are no longer considered to be distinct centres, but rather as part of a larger system (24,27,28).

Specific language regions

Beyond the two traditional language areas, numerous additional regions have been identified as contributing to language processing. These include the right fusiform gyrus, which is activated during the analysis of pictographic languages. The right temporoparietal region exhibits increased activation when Chinese is being spoken and understood (21,29). In addition, the right and left anterior temporal lobes are important in the comprehension of tonal languages (21,29). The anterior supplementary motor area, anterior cingulate cortex, middle frontal gyrus and left caudate nucleus are involved in the switching of languages; while the IFG is only activated when there is a demand for more regions to be activated due to a high amount of information (30-34).

Fibre tracts and language networks

Subcortical fibre tracts in the brain connect the numerous cortical regions responsible for language processing and forming language networks. The language network is best described through the dual-stream model, comprising the ventral and dorsal streams as the two main pathways (35-37). The ventral stream encompasses the extreme fibre capsule system, which further comprises the inferior longitudinal fasciculus, inferior fronto-occipital fasciculus, uncinate fasciculus and middle longitudinal fasciculus. This system connects the middle temporal lobe to the ventrolateral prefrontal cortex, serving a key role in relating incoming auditory information to semantic meaning. Therefore, damage to the ventral stream can lead to impairments in recognition, expression and memory (35,38,39).

The dorsal stream comprises the arcuate fasciculus (AF) and superior longitudinal fasciculus, two major tracts in which auditory information and perception are integrated. These are then used to integrate an expressive response in the superior temporal lobe and premotor cortices in the frontal lobe (35,37). The AF connects the frontal cortex with the occipital and parietal cortices to create expressive responses and damage to the AF may result in deficits in expressive functioning (35-37,40,41).

An understanding of the complex structural and functional organization of language networks is needed when selecting mapping strategies for different circumstances. While the identification of anatomical landmarks offers a general framework, intra- and preoperative techniques are required to fully take inter-individual variability and tumour-induced plasticity into consideration. The subsequent sections will review the principal modalities used to localise and monitor language areas during glioma surgery.

3. Techniques and methodologies for protecting language function areas in brain tumour surgery

DES and SEEG. Intra-operative DES

DES is the most established method for the intraoperative mapping of eloquent cortical regions and subcortical tracts and continues to be in use today (21,42). DES was first described in the 1990s and has become an invaluable tool during surgeries to achieve maximum safe resection of tumours without affecting brain function, including that of language processing. It is extensively used in the mapping of motor and sensory pathways, but also in intraoperative language mapping, where awake surgeries are required (21,42,43).

During awake brain surgery, patients are asked to perform language-associated tasks, such as object-naming and counting and are tested for verbal fluency while cortical regions surrounding the tumour are stimulated. This stimulation of the brain momentarily disrupts the normal functioning of the area, allowing the surgeon to deduce the functions of that region and identify important language regions. A professional specializing in speech and language, such as a speech therapist or neuropsychologist, will often monitor the patient to accurately map out the targeted brain regions (21,44-48).

Awake language mapping procedures now increasingly recognise and utilise DES. DES possess two modalities, both of which identify similar cortical language or motor sites, but differ in their detection of activation zones at the periphery (49). Both offer distinct advantages and their specific usage is based considering the tumour location, resection goals and proximity to eloquent areas. Bipolar DES is established to have a higher special precision and is used in identifying functionally critical areas, such as those in the premotor cortex (8). Unipolar DES delivers high-frequency stimulation trains, provides increased sensitivity for subcortical mapping and can estimate the distance to key tracts based on response threshold (49,50).

The design of language tasks used during intraoperative mapping is just as important. Established paradigms used include object-naming, image description, verb generation, number counting and sentence or syntax completion (51). The importance of task selection in association with lesion topography has been demonstrated in recent studies, whereby noun-naming tasks for temporal lobe tumours and syntactic production tasks for lesions involving the IFG have contributed to a reduction in false negatives and an increase in mapping sensitivity (45,49-52).

The application of DES involves the use of bipolar stimulation with a current ranging from 1.5-6.0 mA and is applied on neural tissue for ~3 sec at a time. Each targeted region is stimulated numerous times to affirm its function. A detected language region is determined when the benchmark of 66% for language task failure is reached (8,45,53).

Despite this, there exists limitations for DES. There is a chance that transient language disturbances observed during stimulation are false positives, resulting from the spread of electrical current through the neural tissue or from patient fatigue. False negatives may also occur, especially when the technique is applied at a suboptimal current intensity or when there is a mismatch between the stimulated region and the task used, highlighting the importance of appropriate task selection in this technique (45).

Due to the complexity of language system and limitations in equipment and resources, there is no universally standardised protocol for language mapping during DES, resulting in notable variability between institutions. Object-naming is commonly used as a measure of language function and is considered the most established method; when choosing tasks, the location of the lesion should be carefully considered to minimise postoperative language function deficiencies (51).

Recent studies have identified motor and non-motor speech arrests caused by DES. Motor speech arrest, identified through the involuntary movement of facial muscles, is the result of DES disrupting the functioning of tissues responsible for motor control. Non-motor speech arrest is the disruption in the transformation of phonological understanding to motor response and primarily occurs in the left IFG (Broca's area) (45,54).

SEEG. SEEG is a technique that complements DES involving the implantation of electrodes in the brain to record electrical activity. This is often paired with DES to determine eloquent areas and is an effective and safe technique that is reliable in determining areas with language functioning. Although it was originally designed for epilepsy surgery and its use outside its original purpose remains limited, further research may lead to its use in tumour resection and intraoperative language mapping, thereby realising its potential in glioma surgery (11,21,55-59). In patients with glioma, SEEG deployment is complicated by tumour-associated mass effect, anatomical distortion and infiltrative growth patterns, which may alter the spatial relationship between cortical and subcortical structures. These factors make electrode placement more challenging compared with epilepsy patients, whose anatomy is relatively preserved. Consequently, SEEG remains largely restricted to selected cases and research settings (56).

A number of preoperative modalities are also used to determine language lateralization and cortical localization in preoperative settings. This includes the Wada test or intracarotid amobarbital procedure, which encompasses the transient inactivation of one of the cerebral hemispheres to determine specific hemispheric dominance in language and memory. Despite this being an invasive procedure, it remains useful under circumstances where non-invasive methods provide inconclusive results or there is lack of patient cooperation. Studies have shown a moderate-to-high consistency between Wada testing and functional MRI (fMRI) in determining language dominance in cerebral hemispheres (9,10).

Non-invasive approaches, such as navigated transcranial magnetic stimulation (nTMS) have also become popular in preoperative functional mapping. nTMS requires the use of repeated magnetic pulses to transiently disrupt local cortical activity. With the use of task-based paradigms, nTMS can also identify eloquent language areas with a high spatial accuracy. Tractography can also be used alongside nTMS to delineate perilesional subcortical language tracts. While useful, nTMS is expensive and requires specialized personnel and protocols (42,49,60).

Magnetoencephalography (MEG)

MEG is an additional preoperative modality capable of high temporal resolution through the detection of magnetic fields produced by neural activity. MEG maps the spatiotemporal dynamics of language processing and despite its spatial resolution being more limited compared with that of fMRI and the technique being more expensive and less widely available, it remains a valuable complementary tool in complex or ambiguous cases (61).

fMRI and DTI. fMRI

fMRI is a widely implicated technology used for mapping cortical areas that are responsible for language functions in a preoperative setting, before a brain tumour resection surgery. Task-based fMRI, which is used to identify language-eloquent regions in the brain, typically utilises language tasks such as image-naming and reading to achieve results (57,62-64). Despite its advantages, fMRI has not yet reached the level of accuracy needed for the most precise cortical mapping, due to the intrinsic complexity of language networks and the phenomenon of neurovascular uncoupling. The decreased level of precision in the ability to pinpoint specific language areas makes fMRI not reliable enough when used on its own (45,62,65,66).

The main downside of fMRI is its dependency on the ability of the patient to perform tasks during the scan, which causes difficulties for patients with cognitive or physical impairments. The nature of the language function is diverse and the cortical areas involved are varied and numerous (45,57,67). Studies have demonstrated the variability of the sensitivity and specificity rates for fMRI, which often depend on the language tasks selected, the strength of the MRI magnet and the analysis models used. This further highlights the importance of having standardised guidelines in the application of fMRI as a cortical mapping tool for language-eloquent areas (45,62,66,68-71).

DTI. DTI is a neuroimaging technique used to map white matter tracts in the brain. It is used to provide anatomical information on the pathways involved in language processing. DTI-tractography is now used widely in tumour resection surgeries, where it is involved in the planning of glioma surgeries, along with providing real-time guidance during the surgery, which is important when the tumour is situated near language tracts (72).

The use of DTI allows for the visualization of how different brain regions demonstrate intricate and complex connectivity, showing how a resection surgery may affect language function. The preoperative use of DTI can predict the location of key tracts, allowing surgeons to plan out their surgery in advance to minimise any impairment dealt to the language functioning of patients. DTI-based neuronavigation provides intraoperative guidance for the surgeon, allowing for further accuracy in the prevention of damage to language-eloquent areas, especially when awake mapping is not possible (73-75). The Gibbs tracking method is one such DTI approach that is noted for its superior reliability in mapping language pathways such as the inferior-occipital fasciculus. Alternative methods, such as streamline propagation and tensor deflection, do not achieve the same level of accuracy as DTI (73,76).

Recent studies have shown that DTI is not only used for surgical planning but also for prognostic modelling of language function. A previous study has showed that the preoperative integrity of key language-related white matter tracts, such as the AF and the inferior fronto-occipital fasciculus, is associated with the baseline language performance and degree of postoperative recovery in patients with glioma (77).

Advanced connectome-based models derived from DTI are able to quantitate structural network efficiency and vulnerability, metrics which are now used in machine learning frameworks to develop personalised neurosurgical predictions of postoperative functional decline or recovery, representing promise for future implementation of individualised medicine (77). The application of DTI has potential in preoperative planning, but also in predicting postoperative outcomes. The preoperative structural integrity measured using DTI is associated with preoperative language function and postoperative recovery. This capability can be improved upon through the pairing of DTI with other neuroimaging techniques, such as fMRI (77).

Combined use of fMRI and DTI. Combination of fMRI and DTI has been shown to enhance the accuracy of preoperative planning in a number of cohort studies, with their combined use providing a more comprehensive representation of language functions and their associated white-matter tracts. This, in turn, improves surgical planning and may facilitate improved postoperative recovery compared with the use of either modality alone (68,69-71,73,76,78-82).

Functional ultrasound (fUS)

fUS is an emerging tool that has shown promise upon use in surgical procedures where the protection of language function is an important goal. This technique uses the ability of an ultrasound to provide anatomical and physiological information during surgery, resolving the issue of poorly defined tumour margins of previous technologies that occur when solely using preoperative imaging techniques (83).

Its capability to detect task-evoked cortical functional responses also makes fUS an important tool for awake brain surgeries. During these surgeries, fUS can monitor changes in cortical activity that are caused by tasks performed by patients and is especially relevant when the tasks are language-based and the tumour is situated near language-processing centres. fUS can also differentiate between tumour and normal brain tissue, providing further intraoperative guidance (83).

B-mode ultrasounds provide a wide range of morphological information, such as size, shape, internal echo texture and spatial structure, that can be paired with fusion imaging to improve image comprehension. Doppler imaging is able to assess both the anatomy and function of vascular networks, while contrast-enhanced ultrasound is used to perform angiosonography, providing further insights into tumour vascularization and perfusion. This not only aids in the detection of tumours, but also the characterization of tumours and identification of residual tumour tissue following resection (84). Elastography is an additional promising ultrasound-based technique that can assist in differentiating lesion grades and the extent of tumour tissues. Multimodal intraoperative ultrasound may be the most effective way to use fUS, allowing the ultrasound to have real-time impacts during the surgery, while being able to synergise with other intraoperative imaging techniques to further improve surgical planning (84). Focused ultrasounds are useful tools for intraoperative use and may be improved upon when couple with other imaging techniques such as DES (85).

Hyperspectral imaging (HSI)

HSI is an emerging technology that demonstrates marked potential for use in brain tumour resection surgeries. HSI has shown promise with regard to the preservation of cortical areas associated with language processing. HSI is a non-contact, non-ionizing and non-invasive imaging technique that provides real-time information on the characteristics of the tissue analyzed, all based on observed tissue properties using the HSI. This allows HSI to be used intraoperatively to identify tumour and healthy brain tissue (86).

The HSI captures a wide range of light across the light spectrum, allowing it to collect data from various wavelengths. These data are then organised and presented as a three-dimensional matrix referred to as a hyperspectral data cube. Both spatial and spectral information is provided, which enables in-depth analysis of the target tissue. Within spectral information, absorption, reflection and scattering properties allow tissue characteristics to be determined. For example, HIS can measure haemoglobin concentration and oxygen saturation, which are primary indicators of angiogenesis or hypermetabolism and are both indicators of tumour tissues (86).

A limitation of HSI is that the high dimensionality of hyperspectral data renders real-time processing challenging. Recent advancements have introduced novel dimensionality reduction schemes and processing pipelines to resolve these issues (86-88). The additional use of T-distributed stochastic neighbour embedding (t-SNE) for dimensionality reduction with semantic Texton Forest is one such approach. This facilitates the generation of detailed tumour classification maps, improving the precision of tumour margins intraoperatively (86).

Advancements in hyperspectral cameras, notably in their miniaturization, have resulted in HSI becoming more accessible and practical. A novel technique known as Lightfield HSI allows for high video retrieval and spectral resolution at the same time. Developments in this field are aiming to achieve true real-time, wide-field and label-free intraoperative tissue identification and differentiation (87). The ability of this technology to be used in conjunction with other imaging techniques has also been explored, namely through Raman spectroscopy. Raman HSI provides non-destructive and chemically rich information regarding target tissues, making this technique invaluable for differentiating tumours from healthy tissue (89).

Currently, HSI is not in widespread use. The main barriers to clinical translation are processing speed, lack of hardware and software standardization and limited multi-centre outcome data. Despite this, HSI represents a promising imaging technique currently in development (90). Furthermore, the successful clinical translation of HSI into routine neurosurgical practices faces a number of challenges. The high dimensionality of hyperspectral data can result in delays in image processing, affecting its intraoperative application (86,87). Modern systems do not possess standardization across institutions, which prevents their smooth integration into surgical workflows. The lack of standardization in calibration and spectral atlases for brain tissues presents problems for reproducibility and cross-platform validation (87,88).

Advances in machine learning-based processing, such as t-SNE or discriminant analysis, are often needed for HSI outputs to be properly interpreted and to distinguish tumour tissue from healthy tissue (86,89). A previous proof-of-concept study has demonstrated the practicality of real-time semantic segmentation, but very few have directly associated HSI-guided resection with improved functional and oncological outcomes (88).

In order to overcome these difficulties, improved hardware optimization, real-time signal processing algorithms and developed integration into neuronavigational systems is needed. Standardised protocols and multi-centre validation studies will also prove key in the integration of this technology into clinical settings (87,88).

Intraoperative optical imaging (IOI)

IOI is a functional imaging technique that is contactless and non-invasive. It has the capabilities to detect metabolic changes, such as cerebral blood volume and oxygenation, by analysing camera images from the exposed brain surface. Cortical optical properties are scrutinised, providing surgeons with key information regarding cortical organization and neuronal connectivity (91). Despite having its uses in cortical mapping, the specificity of the activity maps generated during speech tasks are too low to compete with other techniques, such as DES. Further research is therefore needed to improve IOI (91). A comparison of key techniques for language mapping in glioma surgery is provided in Table I.

Table I.

Comparison of key techniques for language mapping in glioma surgery.

Technique	Application stage	Advantages	Limitations	Clinical outcomes	Typical complications	Post-operative language recovery	(Refs.)
Bipolar DES	Intraoperative	High spatial precision and effective for identifying discrete cortical language areas	Higher risk of after-discharges and seizures	High spatial specificity and sensitive for focal cortical language sites	Lower stimulation thresholds and higher risk of after-discharges in some series (seizure risk manageable)	In awake paradigms, transient errors are common and permanent deficits are relatively low in modern series	(8,94,102)
Unipolar DES	Intraoperative	Broader stimulation field, lower seizure risk and more efficient subcortica l mapping	Less spatially specific	Broader field and is useful for subcortical mapping and probe-to-tract distance estimates	Requires higher currents and some studies report lower after-discharge rate vs. bipolar	Complementary to bipolar and helps reduce missed subcortical eloquent zones when combined	(50,94)
Wada test	Preoperative	Determines hemispheric dominance and is suitable for patients unable to perform tasks	Invasive, exhibits low spatial resolution and is gradually being replaced	Historically established method for hemispheric dominance and is now used selectively	Small procedural risk (vascular complications are rare)	Predicts hemispheric dominance and is not suitable for fine cortical mapping	(9,10)
SEEG	Preoperative and intraoperative assist	Allows mapping of deep structures and is useful in epilepsy or special cases	Highly invasive and currently demonstrates limited glioma-specific usage	Enables mapping of deep/sulcal language areas not reachable by surface DES and is emerging for language mapping	Invasive, exhibits a small risk of haemorrhage/infection and stimulation risks are present	Can predict language decline in selected series and is mainly used in specialised or complex contexts	(56)
nTMS	Preoperative	Non-invasive and demonstrates high accuracy	Expensive and requires specialised personnel and task protocols	Good cortical concordance with DES (≥80% in numerous reports) and is useful for planning	Well tolerated, headaches are common and seizures are rare (<1-2% in cohorts)	Cohort studies report reduced immediate postoperative deficits	(60)
MEG	Preoperative	Excellent temporal resolution and detects fast neural activity	Limited spatial resolution, costly and less widely available	Excellent temporal resolution and complementary cortical localization	Non-invasive and limited availability/cost	Adjunctive role and the independent impact on long-term outcomes is unclear	(61)
fMRI	Preoperative	Non-invasive and good for cortical localization	Susceptible to neurovascular uncoupling and is task-dependent	Useful for lateralization and spatial concordance vs. DES variable (70-90% in many series, depending on task and analysis)	No direct physical risk, but the main issue is false positive or negative localization if used on its own	When combined with intraoperative mapping, supports good recovery, but concordance is moderate when used alone	(62)
DTI	Preoperative and intraoperative navigation	Visualises language-associated fibre tracts and aids surgical path planning	Tractography accuracy decreases near tumours	Improves visualization of tracts and helps reduce inadvertent tract injury when integrated	Reconstruction artifacts and reduced accuracy near infiltrative tumour	Tract integrity is associated with risk of decline and recovery trajectory	(63,77)

fMRI, functional magnetic resonance imaging; DTI, diffusion tensor imaging; nTMS, navigated transcranial magnetic stimulation; MEG, magnetoencephalography; DES, direct electrical stimulation; SEEG, stereo-electroencephalography; HSI, hyperspectral imaging; fUS, functional ultrasound.

4. Current limitations

Limitations to language-mapping technologies that constrain their precision and vulnerability to current practices still exist, despite marked advances in these technologies. The available data varies greatly across mapping methods when it comes to the strength of their data from an evidence-based perspective. DES is supported by a number of cohort studies and meta-analyses, providing the most notable clinical evidence for intraoperative language mapping (11,17,42,44,50,53,92-94). Despite DES being widely regarded as an established method for intraoperative language localization, the variability in stimulation protocols and task batteries present limitations in inter-centre reproducibility and international consensus on stimulation parameters and standardised task sets remains to be decided (11,17,42,44,50,53,92-94).

Small prospective and retrospective series studies are primarily used to validate non-invasive modalities, such as fMRI and DTI, with task design and tumour characteristics driving the heterogenous accuracy (17,87). Techniques such as fMRI and DTI are also dependent on task compliance and are vulnerable to neurovascular uncoupling, particularly in patients with large or infiltrative tumours. Navigated TMS has been assessed in numerous prospective single-centre cohorts with moderate sample sizes, exhibiting high concordance with DES but lacking multi-centre randomised evidence (42,49,60). Furthermore, MEG research remains largely exploratory, with limited patient numbers and variable reproducibility (61). Furthermore, newer methods, such as HSI fUS represent promising task-independent intraoperative modalities that further require improved abilities in real-time processing, standardization and multi-centre validation. While DES currently represents the only technique with robust evidence behind it, the clinical integration of other modalities will depend on further prospective data and standardised protocols (Table II).

Table II.

Evidence-based levels of key techniques for language mapping in glioma surgery.

Technique	Main evidence base	Evidence level
Wada test	Retrospective series and comparative studies vs. fMRI/nTMS	Moderate
fMRI	Retrospective series, small prospective studies	Moderate
DTI	Retrospective and pilot prospective	Moderate
nTMS	Prospective single-centre cohorts	Moderate
MEG	Exploratory studies and small sample size	Low
DES	Large prospective cohorts and meta-analyses	High
SEEG	Case series with small prospective cohorts (primarily epilepsy)	Low
HSI	Proof-of-concept only	Very low
fUS	Pilot studies only	Very low

Evidence levels were assigned by the authors based on the overall strength and consistency of glioma-specific evidence supporting each modality rather than according to a formal evidence-grading system. Classification was determined using the following criteria: i) ‘High’, supported by numerous large prospective cohort studies and/or systematic reviews or meta-analyses with consistent findings in glioma patients; ii) ‘Moderate’, supported by prospective cohort studies or numerous retrospective studies with generally consistent results, but lacking large multicentre validation; iii) ‘Low’, supported mainly by exploratory studies, case series or small cohort studies, with limited sample size and; iv) ‘Very low’, supported only by proof-of-concept studies, pilot investigations or preliminary feasibility data, without clinical validation in glioma populations. Evidence classification also considered study design, sample size, prospective validation, consistency of findings and the availability of systematic reviews or meta-analyses (7,23-27,63,82,84,85,89,90,92-95,97). fMRI, functional magnetic resonance imaging; DTI, diffusion tensor imaging; nTMS, navigated transcranial magnetic stimulation; MEG, magnetoencephalography; DES, direct electrical stimulation; SEEG, stereo-electroencephalography; HSI, hyperspectral imaging; fUS, functional ultrasound.

5. Language mapping strategies for special populations

Paediatric patients

Awake language mapping with preparation and simplified or gamified tasks in children requires compatibility commonly found in older patients (95). Despite current clinical evidence primarily being derived from small series, MRI-derived tractography, electrocorticography or nTMS are suitable alternatives for younger or less cooperative patients (96,97).

Low cognition/aphasia

Standard naming tasks are unreliable for patients with aphasia or cognitive deficits. Adaptive paradigms, such as semantic matching, can improve mapping accuracy (98) and multimodal integration (DES + fMRI/DTI/electrocorticography) can reduce false negatives, while task-independent approaches remain experimental (99).

Multilingual patients

Brains of patients who are multilingual may exhibit distinct networks based on age of language acquisition and level of proficiency, making every functionally relevant language important to map (93). Consultation and detailed preoperative history obtained with bilingual neuropsychologists, along with intraoperative testing, is recommended (100).

6. Emerging directions

Recent studies have explored more integrative and predictive strategies. The combination of fMRI, DTI and nTMS increases the reliability of preoperative functional location. In addition, the use of machine learning and artificial intelligence in conjunction with these imaging modalities is resulting in predictive models for postoperative language prognosis, enabling the determination of individualised risk stratification (49,69-71,73,76,78-82,92).

Semantic task optimization, such as dynamic, patient-specific language paradigms, has gained increasing attention in its ability to enable tumour topography and patient-specific language networks. Efforts have also been made to develop mapping protocols for multilingual populations and to map deeper subcortical tracts with newly improved methods, such as tractography-guided stimulation (18,19,29-33,93,100).

7. Translation to clinical practice

Future technologies and modalities should aim to achieve clinical standardization and integration. In addition to technological standardization, successful clinical translation depends on the integration of complementary mapping modalities into a unified surgical workflow. In current practice, preoperative techniques such as fMRI, DTI and nTMS can be combined to identify language-associated cortical regions and subcortical pathways, while intraoperative DES remains the most established technique for real-time functional validation. Rather than relying on a single modality, a multimodal strategy may allow surgeons to integrate anatomical, functional and electrophysiological information throughout surgical planning and resection. When discrepancies arise between preoperative imaging and intraoperative findings, DES generally provides the most reliable basis for surgical decision-making. Such integrated workflows may improve the balance between maximal safe resection and language preservation and are likely to become increasingly important in personalised glioma surgery. Future work should aim to focus on the seamless integration of stimulation parameters across centres, universal task sets that accommodate a broad range of patients and image-derived predictive models tested and validated in future trials. Techniques such as HSI and fUS must undergo regulatory assessments before their integration into pre-existing neuronavigation methods and practices. Fig. 1 presents a proposed multimodal workflow for language mapping in glioma surgery.

Figure 1.

Proposed multimodal workflow for language mapping in glioma surgery. fMRI, functional magnetic resonance imaging; DTI, diffusion tensor imaging; nTMS, navigated transcranial magnetic stimulation; MEG, magnetoencephalography; DES, direct electrical stimulation; SEEG, stereo-electroencephalography; HSI, hyperspectral imaging; fUS, functional ultrasound; AI, artificial intelligence; IOI, intraoperative optical imaging.

The successful translation of the technologies and methods discussed into current neurosurgical practices will have to depend on the cross-disciplinary collaboration between neurosurgeons, neuropsychologists, biomedical engineers and data scientists. Surgical planning with a data-informed and personalised perspective that uses individual connectomics and functional profiles may reinvent the model of ‘functional preservation’ in glioma surgery. Another important consideration in translating language mapping technologies into clinical practice is the dynamic nature of language organization in patients with glioma. Unlike acute brain injuries, gliomas, particularly diffuse low-grade gliomas, often exhibit slow growth patterns that may permit functional reorganization within cortical and subcortical language networks (65,71). A clinical study of 20 patients with low-grade gliomas (14 of which had a glioma in the left hemisphere near language-associated brain areas) reported postoperative findings also could not excluded perilesional reorganization (65). As a result, language function should not be regarded as a static entity confined to predetermined anatomical regions, but rather as a flexible network that may evolve throughout the disease course (101). This concept has particularly important implications for surgical planning. Patients with slowly growing tumours may benefit from more comprehensive intraoperative language assessment and DES-guided awake mapping, as preoperative localization alone may not fully capture ongoing functional redistribution (20,102). In a clinical series of 51 patients with World Health Organisation-grade II gliomas, 16 patients underwent DES-guided awake surgery, supporting individualized intraoperative language mapping (102). By contrast, rapidly progressing tumours may provide less opportunities for adaptive reorganization, potentially resulting in greater correspondence between preoperative findings and intraoperative functional boundaries. Therefore, future language-preservation strategies should aim to incorporate not only multi-modal functional information but also tumour-specific factors, including growth dynamics and the potential for network plasticity, to support more individualised and adaptive surgical decision-making.

8. Future prospects

Future developments in glioma surgery should retain their focus on precision, personalization and integration. Multimodal approaches that combine anatomical, functional and conenectomic data will likely become the standard for preoperative planning. Intraoperative modalities may begin to include real-time functional imaging such as fUS and HIS in addition to DES. In addition, the use of predictive analytics will should provide individualised risk-benefit profiles for tumour resection, thus advancing surgical decision-making processes. Personalised connectome-guided approaches should shift their focus from ‘avoiding eloquent cortex regions’ towards ‘optimizing network efficiency and recovery prospects’, helping to progress functional preservation. Future developments should adhere to the aim of maintaining an optimal balance between maximal oncological resection and the best possible functional outcomes for each patient. Emerging brain-computer interface technologies that combine high-density cortical electrodes with artificial intelligence-based semantic decoding may enable real-time, task-independent mapping of language networks in future clinical applications. Despite being in their early stages, these methods exhibit notable promise in improving functional preservation in glioma surgery.

Funding Statement

Funding: No funding was received.

Availability of data and materials

Not applicable.

Authors' contributions

XL conducted project administration, methodology and manuscript drafting. JO performed data validation and manuscript reviewing. JW was responsible for study conception and design, provided supervision and conducted both data validation and manuscript reviewing. All authors read and approved the final version of the manuscript. Data authentication is not applicable.

Ethics approval and consent to participate

Not applicable.

Patient consent for publication

Not applicable.

Competing interests

The authors declare that they have no competing interests.