Levels of evidence: Numerous reviews and meta-analyses emphasize high overall correspondence in language laterality derived from task fMRI compared to Wada,[@403924; @403960; @403974; @403975] but each emphasized extensive variability in fMRI practices. A recent Delphi survey highlighted that different tasks for lateralization may be suitable for different questions, and optimal fMRI tasks have not yet been established.[@403966]
Given limited data to quantify clinical levels of evidence for any single task, our approach to creating recommendations was informed by:
  1. Dedicated review of neuroscientific studies engaging selected language processes, and how lateralized those processes are,
  2. A published systematic review assessing the lateralization value of different fMRI tasks,[@403970]
  3. Published data comparing laterality from fMRI tasks to the Wada test (but see section 4.3. for limitations with this approach)
The following recommendations are not intended to be prescriptive. Rather, our aim is to highlight a combination of tasks most likely to provide high sensitivity for language lateralization based on current understanding of language-brain relationships.
Recommendation 1. Establishing language lateralization requires a combination of tasks. Different language processes engage different networks of brain regions, and are lateralized to different extents (Fig 2). Our knowledge of these systems is continually evolving and optimal language task development is an area of active research. We advocate adopting established tasks but also further evaluating newly-emerging tasks. Laterality established through a combined analysis of several fMRI tasks has shown better correspondence to Wada laterality than laterality based on individual tasks analyzed on their own.[@403976]
Recommendation 2: Most everyday ‘language’ requires a combination of phonological, lexico-semantic and syntactic processing (in addition to non-linguistic operations, such as cognitive control and working memory). Using at least one sentence-level task is therefore strongly recommended. When considering approaches to language lateralization with fMRI, an historical tendency has been to categorize tasks into those probing the language ‘production network’ (aka ‘expressive’ tasks), ‘comprehension network’ (aka ‘receptive’ tasks) and ‘semantic network’. Since most fMRI tasks engage several (if not all) of these networks to varying degrees, there are practical and theoretical limits to categorizing tasks as one or the other 58–60. Furthermore, it is the task contrast to a specific control or baseline condition, that enables to study specific language processes (section 2.3.3.2). With this caveat in mind, some specific task approaches to consider include:
  1. Language ‘production’*: Generating or producing language, aside from the articulatory system, involves a network of brain regions including the inferior frontal, supramarginal and superior temporal gyri. A systematic review identified word generation tasks as strongly lateralizing,[@403970] especially when using an appropriate ‘active control’. Among studies comparing laterality derived from fMRI versus Wada, letter fluency (9 studies), category fluency (5 studies) and verb generation tasks (5 studies) have been the most evaluated.[@403960; @403974; @403975] Letter fluency engages the frontal lobe more than category fluency, which instead engages the temporal lobe more (e.g.,[@403980] noting also[@403981]). The choice of preferred task is therefore also influenced by the surgical target. Single word generation tasks produce activation that is more confined to the frontal lobes than sentence generation tasks (which additionally engage wider network regions including the temporal lobes[@403982]), and are therefore useful to consider when lateralization is required in the context of a surgical target in the frontal lobe. Patients can have restricted fluency for reasons other than language impairment[@403983]; verb generation tasks are more likely to sustain patient performance and, thereby, fMRI activation.
  2. Language (particularly sentence) comprehension*: Understanding the meaning of language engages a widespread system including the lateral and ventral temporal lobe and angular gyrus in addition to the inferior frontal cortex 65–67. A systematic review identified sentence comprehension tasks, in particular, to be strongly lateralizing.[@403970] Relatively fewer studies have compared laterality from sentence comprehension fMRI tasks with the Wada test. Three of 5 studies in meta-analyses used visually presented sentences with either reading or a semantic/syntactic decision (both compared to a perceptual control). Auditory versions (2 / 5 studies) generally focused on story listening compared to reversed speech, and tended to emphasize passive listening.[@403975] Comprehension can also be assessed using a sentence completion task, in which participants generate a word to complete a sentence stem; or responsive naming, in which a phrase is heard or read (e.g., “long yellow fruit” and interpreted to generate a name: ‘banana’), thus combining language ‘comprehension’ and ‘production’. Such tasks are useful to include to establish lateralization of the wider temporo-parietal (and temporo-parieto-frontal) networks. Simple sentence processing such as reading or passive listening are less lateralizing than sentence comprehension tasks involving a semantic or syntactic decision[@403970] and, therefore, not recommended for language lateralization, at least when used alone and in standard GLM-type analyses, unless limited patient abilities require it.
  3. Tasks requiring a semantic decision, when paired with an active non-linguistic control task, engage additional regions of the anterior and inferior temporal cortex compared to word production and passive sentence listening tasks. Since the use-case for language lateralization is often temporal lobectomy, a word comprehension task involving a semantic decision is important to engage the anterior and basal temporal cortices. The most extensively studied approach contrasts a semantic decision condition, in which the participant decides if an aurally presented word fits a category (e.g., “found in the USA” and “used by people”), with a nonlinguistic tone perception condition.[@403969] In an alternative approach, pairs of words are presented and patients decide if the words are related (e.g., synonyms). In the visual domain, cued naming with action generation (e.g., pen = ‘writing’) can elicit semantic decisions on single words.[@403970] Evaluations against Wada have mostly involved single word decision tasks, presented either visually or aurally (5 studies[@403975]). In a systematic review of fMRI tasks,[@403970] semantic tasks involving a relatedness decision (e.g., word pairs) were more strongly lateralizing than single word decisions. However, lateralization power of semantic tasks is strongly influenced by choices in the fMRI baseline condition, which should be both active and involve a perceptual decision to best isolate linguistic semantic processing (Fig 5).
Modifications for language lateralization in patients with specific impairments are detailed in the full document (for additional considerations in pediatric patients, please also see section 2.3.6). Summary considerations are:
  • Visual impairments: Some patients are blind/have visual field deficits. Others have better auditory than visual attention, and vice versa. When technically feasible (Appendix C), tasks using the auditory modality are crucial for some patients, and beneficial for others.[@403971; @403972; @403987]
  • If sentence level processing is impaired, word-level tasks‡ are one option. Verb generation or single word comprehension (semantic decision) use frequent external cues (e.g., one item every few seconds), which facilitates performance in patients with selective restrictions in executive functioning. Another approach, depending on the nature of the patient’s difficulties, is to adopt passive story listening / sentence reading, which, however, are less strongly lateralizing in general.[@403970]
  • In patients with a/dyslexia, object naming can engage widely distributed lexico-semantic and phonological processing. However, lateralization in naming tasks depends on what ‘picture naming’ is contrasted against.[@403970] It is essential to include a perceptual control (e.g., involving a visual detection/decision on complex un-nameable images) to isolate semantic and phonological components of naming tasks.
  • In patients with anomia: Some studies[@403988; @403989] have evaluated tasks using image pairs (e.g., an object and a separate written word). Further research is needed to evaluate if such “sentence generation” tasks are easier for patients with naming impairments to perform, but one recent study showed this approach[@403990] may offer better lateralization than single object naming.