Differential cross-modal functional plasticity in deaf individuals: A meta-analysis of neuroimaging studies

Summary

A manifest condition of brain plasticity is cross-modal plasticity in deafness, wherein auditory-deprived cortices are recruited by intact sensory modalities. While cross-modal plasticity is well documented, the underlying principles remain debated, centering on the functional change and functional preservation principles. To quantitatively evaluate these principles, we conducted the largest coordinate-based meta-analysis to date, synthesizing 47 studies (801 deaf and 880 hearing participants) spanning linguistic and non-linguistic tasks. Deaf individuals exhibited consistent cross-modal activation in the bilateral superior temporal gyrus (STG). The left anterior STG was selectively recruited during linguistic tasks, consistent with preserved language function, whereas the left posterior and right STG were engaged across both task types, suggesting a shift toward domain-general visual processing. Cross-modal activation was lateralized, with linguistic tasks engaging the left STG and non-linguistic tasks the right. These findings support a differential cross-modal plasticity framework, wherein functional preservation and change coexist across distinct STG subregions.

Highlights

• •Functional preservation and change coexist across subregions of STG in deaf individuals
• •Linguistic and non-linguistic tasks elicit distinct STG recruitment in deaf individuals
• •Left aSTG selectively engages in linguistic tasks, reflecting preserved language processing
• •Right STG recruited by linguistic and non-linguistic tasks suggests domain-general processing

Biomedical discipline; Complication; Neuroscience

Introduction

Brain plasticity refers to the brain’s capacity to modify its structure and function in response to internal or external changes.^1,2,3 A manifest condition of brain plasticity is cross-modal plasticity of people with sensory deprivation, such as deafness.^4,5 In deaf individuals, the sensory-deprived auditory cortices, particularly the superior temporal gyri (STGs), are recruited during tasks performed by intact sensory modalities (e.g., vision).^6,7

Cross-modal plasticity has been observed across a broad range of visually based tasks in deaf individuals. These tasks include viewing movements,^8,9 perceiving rhythm,^10,11 discriminating facial identity,¹² watching silent films,¹³ reading texts,^14,15 discriminating lip-reading,^16,17 processing sign language,^{18,19,20,21,22} and performing memory tasks.^23,24,25 Moreover, the cross-modal recruitment is linked to improved visual performance^26,27 and can influence auditory rehabilitation outcomes after cochlear implantation.^28,29,30,31

Despite the consensus that cross-modal recruitment occurs in deaf individuals, its underlying principles remain debated. Two major principles have been proposed: the functional change principle and the functional preservation principle.^5,32 The functional change principle suggests that sensory-deprived cortices not only adopt new sensory modalities but also undergo a shift toward entirely new cognitive processing.^33,34 For instance, in deaf individuals auditory regions undergo a functional reorganization that fundamentally alters their computational roles, repurposing them to process visual inputs.³³ In support of this, the STG in deaf individuals is recruited for a wide range of non-auditory tasks, spanning from basic visual perception⁹ to higher-level cognitive functions,³⁵ such as visual working memory—a domain typically supported by the frontoparietal network.^23,24,25

A key question raised by the functional change principle is whether the cross-modal recruited areas reported across studies correspond to the same functional region.^33,36 Specifically, are the newly adopted functions implemented by multiple subregions that all support domain-general processing,^37,38,39 or by different subregions distributed across the STG, each supporting domain-specific processes^17,18? However, this question is difficult to resolve through qualitative review alone due to the heterogeneity across studies in experimental paradigms,³⁴ brain morphology,³³ and activation patterns.^40,41

In contrast, the functional preservation principle posits that auditory-deprived cortices retain their original computational roles, but apply them to inputs from an alternative sensory modality.^33,42 According to this view, auditory experience is not a prerequisite for the emergence of functional specializations in temporal regions, as long as the functional-specific information is available through another sensory modality.⁴³ The functional preservation may occur through the potentiation or unmasking of pre-existing cortical connections.^32,44 Supporting this account, the temporal regions in deaf individuals maintain functional specializations during visual tasks, that are similar to those in hearing individuals during auditory tasks, such as motion perception,⁸ rhythms processing,^10,11,33 and identity discrimination.¹²

The functional preservation principle raises important questions about the extent and prerequisites of preserved functions.³³ Which functions are maintained after auditory deprivation, and which are not⁴³? This issue is crucial for understanding how cortical specializations are shaped by auditory-specific experience⁴³ and for informing the design of effective clinical interventions following cochlear implantation.^27,30

Recently, a unified perspective has been proposed, suggesting that functional change and preservation are not mutually exclusive but rather co-exist within the brain.^33,36 Supporting this view, studies have shown that cross-modal recruitment in the left STG is shaped by linguistic experience, such as the use of sign language, while recruitment in the right STG is primarily driven by auditory deprivation.^17,18,19 Furthermore, Zimmermann et al. demonstrated that the same region of the right STG exhibited both domain-specific recruitment during rhythm discrimination and non-specific responses during spatial perception in visual modality. However, the validity of this perspective still requires more quantitative evidence. A comprehensive quantitative meta-analysis is therefore essential to synthesize disparate findings and identify consistent patterns of cross-modal recruitment.³⁵

To quantitatively evaluate the prevalence of functional change and functional preservation in deaf individuals, we conducted the largest activation likelihood estimation (ALE) meta-analysis to date, synthesizing data from forty-seven neuroimaging studies. First, we examined cross-modal recruitment in auditory cortices by identifying clusters that were more active in deaf versus hearing participants across all the studies. Next, we quantified the relative extent of each principle by categorizing studies into two types of tasks: (i) tasks involving linguistic stimuli (e.g., sign language), probing domain-specific language processing; and (ii) tasks with non-linguistic stimuli (e.g., moving dots), which probe a broad range of visual processes in the absence of linguistic input. We chose this taxonomy because (i) language studies dominate existing neuroimaging literature in deaf (49%), ensuring adequate power, and (ii) language relies on well-characterized, left-lateralized networks, enabling clear inferences about functional selectivity. We then conducted separate ALE meta-analyses for linguistic and non-linguistic tasks to localize their cross-modal recruitment. Pairwise ALE contrasts isolated task-unique activations, and conjunction analyses identified regions common to both domains. We also calculated a laterality index for each cluster to compare hemispheric asymmetries between tasks. Finally, we performed five control analyses addressing potential confounds.

Accordingly, we proposed three hypotheses. First, if functional change governs cross-modal plasticity, both linguistic and non-linguistic tasks would engage identical auditory subregions, potentially with similar hemispheric lateralization. Second, if functional preservation governs cross-modal plasticity, we would expect the two task types to recruit distinct auditory subregions with differing lateralization. Third, if both principles coexist, we anticipate a hybrid pattern, where some subregions are activated by both task types, while others are selectively activated by linguistic tasks.

Consistent with the third hypothesis, we observed a differential cross-modal plasticity pattern in which both recruitment patterns coexist, exhibiting distinct dominance across subregions of the STG. Specifically, linguistic tasks selectively recruited the left anterior STG and demonstrated left-lateralized cross-modal recruitment, suggesting the preservation of its specialized linguistic role despite auditory deprivation. Both linguistic and non-linguistic tasks recruited the left posterior STG and right STG, whereas non-linguistic tasks were associated with right-lateralized cross-modal recruitment, suggesting functional reorganization toward domain-general visual processing. These findings extend the current understanding of cross-modal plasticity, highlight the interplay between cortical architecture, sensory deprivation, and language experience.

Results

Overview of included studies

We included forty-seven studies (Figure 1A, detailed information in Table S1), including 801 deaf participants (mean age, 28.8 ± 8.09 years, 404 females) and 880 hearing participants (mean age, 28.3 ± 6.62 years, 488 females) (Figures 1B–1D). Among deaf participants, 85% had severe-to-profound hearing loss, 88% had lost hearing before the age of 3, 91% were signers, and 9% relied on lip-reading (Figure S1). A total of 11 ALE analyses were conducted, as detailed in Table 1).

Figure 1.

Demographics and overview of the meta-analysis

(A) PRISMA flow diagram for the procedure of study selection.

(B) Distribution of sample sizes in all the included studies (red represents deaf participants, blue represents hearing participants).

(C) Distribution of participants’ age.

(D) Distribution of gender and sign language proficiency.

(E) Word cloud illustrating the most frequent stimuli and tasks.

(F) Foci exhibiting greater activation in deaf participants (deaf > hearing) and hearing participants (deaf < hearing). Abbreviations: ALE, activation likelihood estimation; fMRI, functional magnetic resonance imaging; PET, positron emission tomography.

Table 1.

Descriptive information for all ALE analyses

Contrast	Number of experiments	Number of deaf participants	Number of hearing participants	Number of foci
All the tasks: deaf > hearing	45	758 (female = 382)	837 (female = 471)	324
All the tasks: deaf < hearing	23	347 (female = 182)	387 (female = 211)	175
Hearing only: auditory > baseline	5	/	101 (female = 55)	45
Linguistic tasks (based on stimulus): deaf > hearing	28	448 (female = 235)	447 (female = 255)	207
Non-linguistic tasks (based on stimulus): deaf > hearing	22	403 (female = 221)	484 (female = 279)	126
Dynamic linguistic tasks (based on stimulus): deaf > hearing	20	297 (female = 141)	304 (female = 173)	134
Dynamic non-linguistic tasks (based on stimulus): deaf > hearing	11	147 (female = 89)	204 (female = 120)	72
Linguistic tasks (based on task requirements): deaf > hearing	32	535 (female = 256)	575 (female = 324)	213
Non-linguistic tasks (based on task requirements): deaf > hearing	15	265 (female = 132)	316 (female = 187)	101
Linguistic tasks (based on stimulus): prelingual deaf > hearing	23	351 (female = 175)	386 (female = 213)	140
Non-linguistic tasks (based on stimulus): prelingual deaf > hearing	15	289 (female = 144)	368 (female = 208)	64

Abbreviations: /, not applicable.

Across all studies, 49% used linguistic stimuli, 38% used non-linguistic stimuli, and 13% used both types of stimuli (Figure 1E). The five most common stimuli were sign, word, sentence, gesture, and point-light. The five most frequent tasks included detection, visual judgment, semantic processing, passive viewing, and discrimination.

In total, 324 foci exhibited increased activation in deaf participants compared to hearing participants, while 175 foci showed reduced activation in the deaf group (Figure 1F).

Cross-modal recruitments in auditory-deprived cortex

Across all the forty-seven studies, deaf participants consistently exhibited greater activation than hearing participants in the bilateral temporal cortex, spanning from anterior to posterior STG (BA 22/41, Figure 2A and Table 2). These clusters were identified as cross-modal recruited areas in deaf participants. In contrast, deaf participants exhibited reduced activation in the left inferior occipital gyrus (BA 19) compared with hearing participants (Figure 2B and Table 2).

Figure 2.

Cross-modal recruitment of the auditory-deprived cortex in deaf participants

(A) Greater activation in deaf than hearing participants (deaf > hearing).

(B) Lower activation in deaf than hearing participants (deaf < hearing).

(C) The auditory areas activated in hearing participants during auditory tasks (hearing: auditory > baseline).

(D) The overlap of deaf > hearing (in red) and hearing: auditory > baseline (in blue). The green circle highlights the functional masks of the auditory cortex from Neurosynth. The results are thresholded at p < 0.001 at the voxel level and p < 0.05 at the cluster level.

Table 2.

Regions exhibited altered activation in deaf participants, and auditory regions in hearing participants

Cluster	Hemisphere	Areas	BA	Coordinate (MNI—X Y Z)			ALE value	Z value	Size (mm3)
deaf > hearing
2	left	aSTG	22	−6	−12	0	0.046	8.00	9904
		pSTG	22	−5	−32	8	0.036	6.81
1	right	aSTG	22	60	−4	−6	0.028	5.66	10984
		mSTG	41	62	−20	4	0.056	9.02
		pSTG	41	54	−30	8	0.036	6.82
deaf < hearing
1	left	IOG	19	−4	−82	−4	0.020	5.29	864
auditory > baseline (hearing only)
1	left	mSTG	22	−6	−24	2	0.015	5.41	2360
2	right	aSTG	22	56	−18	−2	0.012	4.65	1272
		mSTG	22	62	−28	2	0.008	3.78
3	right	pSTG	41	50	−34	10	0.012	4.59	944

Abbreviations: BA, Brodmann area; MNI, Montreal Neurological Institute; ALE, activation likelihood estimation; aSTG, anterior superior temporal gyrus; pSTG, posterior superior temporal gyrus; mSTG, middle superior temporal gyrus; IOG, inferior occipital gyrus.

In hearing participants, auditory tasks elicited increased activation in the bilateral STG (BA 22/41, Figure 2C and Table 2). Critically, 82% of the voxels in the cross-modally recruited areas of deaf participants overlapped with the joint auditory mask, with 81% overlapping in the left hemisphere and 83% in the right hemisphere (Figure 2D), indicating consistent cross-modal recruitment of the STG.

Coexistence of language-specific and domain-general cross-modal recruitment

Across all the linguistic tasks, deaf participants exhibited greater activation than hearing participants in the left temporal cortex, spanning from left anterior to posterior STG, and right STG (Figure 3A and Table 3). In contrast, deaf participants exhibited reduced activation in the left angular gyrus (BA 39) and posterior STG (BA 22/39) compared with hearing participants (Figure S2 and Table S3).

Figure 3.

Language-specific and domain-general cross-modal recruitment of the auditory-deprived cortex in deaf participants

(A) Cross-modal recruitment in deaf participants during linguistic tasks (linguistic task: deaf > hearing).

(B) Cross-modal recruitment in deaf participants during non-linguistic tasks (non-linguistic task: deaf > hearing).

(C) The overlap of cross-model recruitment during linguistic tasks (red) and non-linguistic tasks (blue), language-specific cross-model recruitment (pink), and domain-general cross-model recruitment (yellow). The results are thresholded at p < 0.001 at the voxel level and p < 0.05 at the cluster level.

Table 3.

Specific and common cross-modal activation for linguistic and non-linguistic tasks

Cluster	Hemisphere	Areas	BA	Coordinate (MNI—X Y Z)			ALE value	Z value	Size (mm3)
Linguistic task: deaf > hearing
1	left	aSTG	22	−60	−12	0	0.040	7.72	7568
		pSTG	22	−56	−30	6	0.026	5.81
			41	−48	−34	10	0.016	4.23
		MTG	22	−62	−40	10	0.017	4.43
2	right	aSTG	22	64	−10	0	0.020	4.92	6520
			22	62	−6	−4	0.016	4.19
		mSTG	41	56	−30	8	0.024	5.55
Non-linguistic task: deaf > hearing
2	left	pSTG	22	−62	−24	4	0.016	4.53	2928
			22	−52	−34	6	0.017	4.77
		MTG	22	−64	−34	6	0.014	4.12
1	right	aSTG	22	64	−18	0	0.038	7.85	6992
			22	60	−2	−6	0.015	4.40
		mSTG	41	50	−26	4	0.019	5.17
Linguistic task > non-linguistic task
1	left	aSTG	22	−53	−14	4	/	3.72	424
		TTG	41	−56	−14	10	/	3.54
Linguistic task ∩ non-linguistic task
2	left	mSTG	22	−62	−24	4	0.016	/	1744
		pSTG	22	−52	−32	6	0.017	/
1	right	aSTG	22	60	−4	−6	0.013	/	4224
		mSTG	42	50	−26	4	0.019	/
			22	66	−28	10	0.016	/

Abbreviations: BA, Brodmann area; MNI, Montreal Neurological Institute; ALE, activation likelihood estimation; aSTG, anterior superior temporal gyrus; pSTG, posterior superior gyrus; MTG, middle temporal gyrus; mSTG, middle superior temporal gyrus; IOG, inferior occipital gyrus; TTG, transverse temporal gyrus; and /, not applicable.

Across all the non-linguistic tasks, deaf participants showed greater activation than hearing controls in the left posterior STG and right STG (Figure 3B and Table 3). Deaf participants did not exhibit reduced activation compared with hearing participants during these tasks.

The contrast analysis (linguistic task _{deaf > hearing} > non-linguistic task _{deaf > hearing}) revealed language-specific cross-modal recruitment in the left anterior STG and Heschl’s gyrus (Figure 3C and Table 3). No significant clusters were identified in the reverse contrast.

The conjunction analysis (linguistic task _{deaf > hearing} ∩ non-linguistic task _{deaf > hearing}) identified overlapping cross-modal recruitment in the left posterior STG and right STG (Figure 3C and Table 3), suggesting that these regions support domain-general visual processing in deaf individuals.

Left-hemisphere dominance for language tasks in auditory-deprived cortex

Laterality analysis revealed significant left-lateralized cross-modal recruitment during linguistic tasks (AveLI = 0.28, p < 0.001) and significant right-lateralized recruitment during non-linguistic tasks (AveLI = −0.75, p < 0.001). These results suggest that the typical left-hemispheric dominance for language processing is preserved in the auditory-deprived cortex.

Control analysis

Control for stimulus motion confounds

To account for potential confounds related to stimulus motion, we repeated the analyses using only studies that used dynamic stimuli and replicated the main findings. The left anterior STG showed language-specific cross-modal recruitment, while the right STG supported domain-general visual processing (Figure 4A and Table S4). Lateralization analysis further revealed significant left-lateralized activation during dynamic linguistic tasks (AveLI = 0.23, p < 0.001).

Figure 4.

Results of control analyses

(A) The overlap of cross-modal recruitment in deaf participants during dynamic linguistic tasks (red) and dynamic non-linguistic tasks (blue).

(B) The overlap of cross-modal recruitment in deaf participants during linguistic tasks based on task requirements (red) and non-linguistic tasks based on task requirements (blue).

(C) The overlap of cross-modal recruitment in prelingual deaf participants during linguistic tasks (red) and non-linguistic tasks (blue). The results are thresholded at p < 0.001 at the voxel level and p < 0.05 at the cluster level.

Control for stimulus—Task dissociation

To address discrepancies between stimulus types and task demands, we reclassified studies based on task requirements rather than stimulus and replicated the main findings. The left anterior STG showed language-specific cross-modal recruitment, while the right STG exhibited domain-general visual processing (Figure 4B and Table S5). Lateralization analysis further revealed significant left-lateralized activation during linguistic tasks (AveLI = 0.18, p < 0.001).

Control for variability in the onset of hearing loss

To control the potential variability of hearing loss onset on the ALE results, we restricted our analyses to studies involving only prelingual deafness and then replicated the main findings. The left anterior STG exhibited language-specific cross-modal recruitment, while the right STG showed domain-general visual processing (Figure 4C and Table S6). Lateralization analysis further revealed significant left-lateralized activation during linguistic tasks (AveLI = 0.42, p < 0.001).

Robustness test of results

The leave-one-out jackknife analysis confirmed that all clusters identified in the main analyses were robust to the exclusion of individual studies (Figure S3 and Table S7), with an average robustness of 99% (range, 95%–100%).

Publication bias assessment of results

The FSN analysis indicated that most clusters identified in the main analyses were robust to the inclusion of null studies (Figure S4 and Table S7). The only exception was the left inferior occipital gyrus (BA 19), which showed reduced activation in deaf participants compared to hearing controls and had an FSN score of zero, suggesting susceptibility to publication bias. In contrast, clusters exhibiting greater activation in deaf participants than in hearing controls across all tasks showed high FSN scores. For both linguistic and non-linguistic subgroup analyses, all FSN scores exceeded the predefined threshold of 30%, indicating that the results were robust and reliable.

Collectively, these results demonstrate that our findings are robust to potential confounds related to stimulus motion properties, task classification criteria, variability in participant characteristics, the influence of a small number of outlier studies, and publication bias.

Discussion

We addressed theoretical debates regarding the cross-modal plasticity in the auditory-deprived cortex^5,36 through a systematic coordinate-based meta-analysis of neuroimaging studies on deaf individuals. Across all forty-seven studies, deaf individuals consistently exhibited increased cross-modal activation in bilateral auditory cortices, spanning from anterior to posterior STG (BA 22/41/42). Contrast analysis revealed that the left anterior STG showed increased activation exclusively during linguistic tasks (e.g., sign language), consistent with preserved language-related functions. Meanwhile, conjunction analysis identified increased cross-modal activation in the left posterior STG and right STG during both linguistic and non-linguistic tasks, suggesting reorganization toward domain-general visual processing. Furthermore, linguistic tasks elicited left-lateralized cross-modal responses at the STG, whereas non-linguistic tasks were associated with right-lateralized cross-modal responses at STG in deaf individuals. The hybrid pattern of language-specific and domain-general recruitment supports the third hypothesis of differential cross-modal plasticity—and aligns with the view that functional preservation and change coexist in distinct subregions of the auditory-deprived cortex.^11,33,36

Consistent cross-modal recruitments in auditory-deprived cortex

Our results demonstrate that auditory deprivation alters the sensory preferences of auditory regions, consistent with well-documented evidence of cross-modal recruitments in these areas.^18,35,45,46 Notably, the vast majority of studies examining cross-modal recruitment have relied solely on anatomical landmarks to define auditory regions, with only 11% functionally localizing the auditory regions in hearing participants.^{8,10,12,47,48,49} We found that deaf individuals consistently showed greater activation in a broad region of the bilateral STG compared with hearing participants, with 82% of these cross-modally recruited voxels located within the functional masks of the auditory cortex. These findings align with single-unit recordings in the auditory cortex of deaf ferrets⁵⁰ and mice,⁵¹ which show a marked reduction in unimodal auditory neurons and an increase in neurons responsive to non-auditory signals.

Importantly, our results provide quantitative evidence for a hybrid spatial pattern of cross-modal plasticity in the auditory-deprived cortex. Specifically, different subregions of STG appear to exhibit different types of cross-modal recruitment. The left anterior STG exhibited higher activation in deaf individuals than hearing individuals during linguistic tasks (e.g., sign and lipreading), suggesting preserved linguistic functions adapted to visual inputs. In contrast, the right STG exhibited higher activation in deaf individuals across both linguistic and non-linguistic tasks, suggesting recruitments in domain-general visual processing. This asymmetrical spatial organization of cross-modal plasticity aligns with previous observations that auditory deprivation and language experience may differentially affect the left and right STG.^18,52

Similar patterns have been observed in blind individuals. In these participants, the fusiform face area^53,54 and the visual word form area^55,56 showed domain-specific cross-modal recruitment, maintaining their functional specialization via tactile or auditory inputs. In contrast, the primary visual cortex exhibited domain-general cross-modal responses⁵⁷ and was recruited across a wide range of tasks, including Braille reading,⁵⁵ speech processing,⁵⁸ and mathematical calculation.⁵⁹

Functional preservation in the auditory-deprived cortex

During linguistic tasks, we observed robust cross-modal activation at the left anterior STG in deaf individuals, a core area of the left lateralized language network in hearing individuals.^60,61 This observation is consistent with previous results. For instance, early deaf signers recruited left STG more strongly than late-deaf signers⁶²; when tested by phonology and visuospatial matching tasks with the same stimuli, deaf individuals only recruited the left anterior STG activation under the phonology task.⁵² Furthermore, our control analysis confirms that the cross-modal recruitment of left anterior STG depends on linguistic processing per se, not the dynamic features of sign or facial articulations, underscoring its language-specific recruitment.^18,63,64

The linguistic-specific cross-modal recruitment may arise from the early language experience, such as sign language and lip reading, which provide functional scaffolds for the development of language function in these areas.^18,27 Deaf signers exhibited greater left anterior STG activation than both deaf non-signers when passively observing,⁶⁵ comprehending signs,^20,66,67 and speechreading.⁶⁸ Furthermore, deaf signers activate the same left-lateralized fronto-temporal network as hearing individuals⁶⁹ and exhibit stronger functional coupling between left anterior STG and other language areas,⁷⁰ as well as a more pronounced left lateralized activation during reading, compared with hearing individuals.^71,72

It is worth noting that domain-specific recruitment is neither confined to the language processing nor restricted to the left anterior STG.^36,43 Comparable preservation has been observed in temporal regions supporting motion perception,⁸ identity recognition,¹² and rhythm discrimination^10,11 in different areas of temporal cortex. A common property of these preserved functions is that they can be implemented via multisensory inputs and optimized through multisensory integration (e.g., identity recognition, which naturally operates across multiple sensory modalities).^43,44,73 These multisensory integration properties may allow certain temporal regions to maintain their functional roles by strengthening or unmasking the functional/structural connectivity between auditory regions and corresponding multisensory or visual cortices^12,69 after hearing loss.

Functional change in the auditory-deprived cortex

Unlike the left anterior STG, which preserves its core linguistic role, the left posterior STG and right STG are robustly recruited across both linguistic and non-linguistic tasks, suggesting a functional shift toward domain-general visual processing in deaf individuals.³³ This shift aligns with prior studies demonstrating cross-modal engagement of the right STG in a broad range of domains, such as object matching,^19,52 motion processing,³⁹ spatial perception,¹¹ selective attention,⁷⁴ and working memory.^23,25 Non-linguistic tasks (e.g., dot-motion matching) preferentially engaged the right STG, whereas linguistic tasks (e.g., sign processing) more strongly activate its left counterpart.^17,19 Notably, Cardin et al. demonstrated that both deaf signers and non-signers recruited the right STG to sign stimuli, but only signers recruited the left STG, implying that cross-modal recruitment in the right STG stems from auditory loss.

These patterns of domain-general recruitment may be associated with the functional versatility and/or limited multisensory integration capacity of these regions in hearing people. The right STG, especially its middle and posterior sections, is involved in a wide range of auditory processes, including acoustic-phonetic feature encoding,⁷⁵ prosody perception,^76,77 and spatial processing.^78,79 This multifunctional profile may contribute to their functional flexibility after auditory deprivation.⁴³

In addition, the observed cross-task recruitment may also reflect non-specific cognitive processes that are shared across different tasks. Consistent with this interpretation, studies in blind individuals reported that cross-modal recruitment of the deprived visual cortex was associated with higher-order cognitive processes,^57,80 such as attentional engagement,⁸¹ executive control,⁸² and preparatory activity.⁸³ However, in coordinate-based meta-analyses, systematically investigating these factors is difficult. Further studies with direct experimental manipulations are needed to disentangle effects of these non-specific cognitive processes.

Functional plasticity beyond the auditory cortex

Although our results highlight cross-modal plasticity in the bilateral STG, functional plasticity in deaf individuals extends beyond the auditory regions and language domains. In the visual domain, deaf individuals exhibited similar dorsal and ventral streams as hearing individuals, but with reduced activation in the right parietal lobe and increased activation in the left inferior parietal lobe.⁸⁴ Likewise, numerical cognition in deaf individuals recruited the canonical frontoparietal network for quantity estimation,⁸⁵ arithmetic,⁸⁶ and geometry,⁸⁷ but with increased activation in parietal regions and decreased activation in the left inferior frontal gyrus.^86,88 In the domain of executive control, such as task switching³⁵ and working memory,⁸⁹ deaf individuals showed increased activation in temporal regions and stronger connectivity between the temporal and frontal areas, accompanied by superior performance on cognitively demanding tasks.^24,25

Collectively, these observations reveal a regionally differential functional plasticity across the cerebral cortex following auditory deprivation. Specifically, frontoparietal regions involved in high-level cognitive functions, such as theory of mind, arithmetic, and memory, remain largely functionally unaffected.³⁷ Temporal regions supporting supramodal or multisensory computations, such as motion perception,⁸ identity recognition,¹² and speech perception,^46,47 tend to preserve their original functions via alternative sensory inputs. In contrast, temporal regions characterized by greater functional versatility and/or limited multisensory integration capacity tend to adopt new roles, such as domain-general visual processing.^18,39,52

Future direction

These differential patterns of cross-modal plasticity raise two critical questions. First, to what extent do distinct plasticity patterns account for cognitive deficits and compensatory abilities in deaf individuals^6,26? Second, following hearing restoration (e.g., cochlear implantation), how do these patterns of plasticity contribute to or predict recovery in auditory processing, audiovisual integration, and speech communication^29,90,91? One hypothesis is that functionally preserved regions act as scaffolds, supporting auditory and speech rehabilitation following cochlear implantation. For instance, activity in several cross-modally recruited areas has been found to positively predict auditory and speech outcomes.^28,92 Conversely, regions that have undergone functional change may be associated with maladaptive outcomes.⁹³ Therefore, targeted auditory training tailored to specific functions may be required to reverse cross-modal recruitment in functionally changed regions and restore their original roles.

To address these questions, future studies could focus on longitudinal research tracking cross-modal recruitment before and after cochlear implantation,²⁹ investigating how both functional preserved and changed regions contribute to auditory and speech recovery over time. Integrating multimodal neuroimaging techniques with broader behavioral measurements across large sample sizes⁹⁴ will provide a clearer understanding of the relationship between cross-modal recruitment, sensory experience, and functional compensation or deficits. Additionally, research that examines individual differences, such as age of implantation, duration of auditory deprivation, and pre-existing language proficiency, may offer a more nuanced perspective for developing individualized approaches to sensory and speech recovery.

Limitations of the study

While this study revealed robust and consistent cross-modal recruitment in deaf individuals, several limitations should be acknowledged. First, the coordinate-based meta-analyses with functional magnetic resonance imaging (fMRI) data identify converged BOLD signals but lack the resolution to map underlying computational architectures. The observed spatial patterns may be insufficient to delineate fine-grained anatomical or functional segregations within the temporal cortex. Moreover, the lack of access to raw fMRI data precluded the use of more sophisticated multivariate approaches, such as representational similarity analysis, which could further elucidate the brain representations of linguistic versus non-linguistic processing.

Second, our results focus on spatial convergence and do not provide direct evidence linking cross-modal recruitment in the STG to behavioral or cognitive performance. Although previous studies reported that left STG recruitment correlates with compensatory language enhancement in deaf individuals,^7,26,27,95 the functional significance of these plastic changes remains to be explicitly tested in future integrated studies.

Third, to maintain sufficient statistical power for the ALE analysis, we utilized a broad linguistic/non-linguistic taxonomy rather than granular cognitive domains. This categorization may mask variances in top-down modulation, such as cognitive load, executive control, and attentional demands. Thus, the differences between linguistic and non-linguistic tasks may partly reflect task-general attentional mechanisms rather than purely domain-specific processing. Future research employing parametric manipulations of cognitive demand is essential to disentangle these factors.

Finally, the control analysis for dynamic non-linguistic tasks was based on a limited number of studies (n = 11), falling below the recommended threshold (n = 17) for robust ALE inference.⁹⁶ These findings should be interpreted with caution until validated by a larger body of literature.

Conclusion

We conducted a comprehensive meta-analysis to address ongoing theoretical debates regarding cross-modal plasticity in the auditory-deprived cortex. Our findings support a differential cross-modal plasticity pattern in which both recruitment patterns coexist, exhibiting distinct dominance across bilateral STG. The left anterior STG showed selective recruitment for language tasks, suggesting the preservation of its specialized linguistic role despite auditory deprivation. In contrast, the left posterior STG and right STG responded broadly to diverse visual inputs, suggesting functional reorganization toward domain-general visual processing. Moreover, linguistic tasks elicited left-lateralized cross-modal recruitment, whereas non-linguistic tasks were associated with right-lateralized cross-modal recruitment, highlighting a hemispheric dissociation between preserved and repurposed function. Our findings extend current understanding of cross-modal plasticity in deaf brains, highlight the interplay between cortical architecture, sensory deprivation, and language experience.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Chenjie Dong (dong_chenjie@m.scnu.edu.cn).

Materials availability

This study is a meta-analysis, which does not involve any reagents or new materials.

Data and code availability

• •Detailed information for all studies included in the meta-analysis is presented in Table S1.
• •All data and original code have been deposited in the Open Science Framework (OSF) at https://osf.io/nzqvp/overview?view_only=4da4b279b91844d8be564833b2c1f381 and are publicly available as of the date of publication.

Acknowledgments

This research was supported by the National Natural Science Foundation of China (32500931; 32171051), the Key Research and Development Program of Guangdong (2023B0303010004), and the Research Center for Brain Cognition and Human Development, Guangdong (2024B0303390003). We extend our heartfelt thanks to the deaf community in Guangdong Province, China, for their ongoing support and inspiration.

Author contributions

X.Z., conceptualization, data analysis, writing manuscript, and revising manuscript; G.L., Z.W., and P.P., data analysis, results visualization, and revising manuscript; C.D., conceptualization, data analysis, writing manuscript, revising manuscript, supervision, funding acquisition; S.W., conceptualization, revising manuscript, supervision, funding acquisition, project administration, and resources.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
Neurosynth	Neurosynth Organization	RRID:SCR_006798
Software and algorithms
GingerALE version 3.0.2	BrainMap	RRID:SCR_014921
xjView	AliveLearn	RRID:SCR_008642
FreeSurfer	Martinos Center for Biomedical Imaging	RRID:SCR_001847
PySurfer	PySurfer Development Team	RRID:SCR_002524

Method details

Study identification

Figure 1A illustrates the literature search and study selection procedures. We conducted and reported the meta-analysis in accordance with current guidelines for neuroimaging meta-analyses⁹⁶ and the updated PRISMA framework.⁹⁷ The search was restricted to studies involving human participants and published in English. The following keywords were used: [fMRI OR PET] AND [deaf OR deafness OR hearing loss OR hearing deprivation]. A systematic search was conducted across PubMed, Google Scholar, and Web of Science for publications from January 1990 to July 2024, yielding eight hundred and eighty-seven articles.

Study screening

After removing duplicate records, three authors (XZ, ZW, and CD) independently screened four hundred and twenty-seven unique articles. Discrepancies were resolved through discussion in regular meetings.

The inclusion criteria for screening were as follows: (ⅰ) The study used task-based fMRI or PET; (ⅱ) included both deaf and hearing participants; (iii) reported group-level results based on whole-brain analyses (not ROI-based or multivariate analyses); (iv) reported inter-group activation differences and provided stereotactic coordinates; (v) results were reported in standardized coordinate space: Montreal Neurological Institute (MNI) or Talairach.^98,99

The exclusion criteria for screening were as follows: (ⅰ) Participants had unilateral deafness; (ii) participants with age-related hearing loss; (iii) participants with cochlear implants; (iv) studies focused on non-visual modalities (e.g., tactile or olfactory tasks); (v) case reports.

According to the predefined criteria, three hundred and eighty articles were excluded for the following reasons: absence of an experimental task (n = 102); lack of a hearing control group (n = 26); lack of whole-brain activation results (n = 42); lack of between-group comparisons (n = 7); absence of significant group differences (n = 5); missing coordinates for between-group contrasts (n = 10); inclusion of participants with unilateral deafness (n = 48); focus on age-related hearing loss (n = 51); participants with cochlear implants (n = 76); focus on non-visual modalities (n = 3); case reports (n = 9); and incorrect stereotactic coordinates (n = 1).

Ultimately, forty-seven studies were included in the final dataset (Table S1). Of these, forty-five reported greater activation in deaf participants compared to hearing participants, while twenty-three reported lower activation in deaf participants.

Data extraction

We extracted the following information from each of the forty-seven studies (Table S1): (i) publication details, including title, journal, year, and first author; (ii) participant demographics (Figures 1B–1D), such as sample size, mean age, gender, age of hearing loss onset, and degree of hearing loss; (iii) experimental variables (Figure 1E), including stimulus content, stimulus modality, and task type; and (iv) results variables (Figure 1F), including contrast types, normalization space, and peak activation coordinates.

Quantification and statistical analysis

Activation likelihood estimation

Activation likelihood estimation (ALE) analysis is the most widely used algorithm for coordinate-based meta-analyses in neuroimaging research.¹⁰⁰ It identifies spatial convergence of activation peaks across experiments by comparing the observed overlap to a null distribution generated through random spatial association.^101,102 In ALE analyses, each reported peak coordinate is modeled as the center of a 3D Gaussian probability distribution, with the width determined by the sample size of the original study. A modeled activation (MA) map is computed for each experiment, representing the likelihood of activation at each voxel. These MA maps are then aggregated across studies to produce a group-level ALE map, reflecting regions of consistent activation across experiments. Statistical significance is assessed by comparing observed ALE values to a null distribution generated by randomly sampling one voxel per MA map and computing ALE scores in the same manner.

All analyses were performed using GingerALE version 3.0.2 (https://brainmap.org/ale). To ensure coordinate consistency, Talairach coordinates were converted to MNI space using the icbm2tal transformation function.¹⁰³ A less conservative anatomical mask (dilated by 2 mm) was applied during analysis. Significant clusters were determined by comparison to a null distribution generated from 1000 simulated datasets with randomly distributed foci. A voxel-level threshold of p < 0.001 and a cluster-level threshold of p < 0.05 were used for multiple comparisons correction.^96,104,105 All the results were initially visualized with xjView (https://www.alivelearn.net/xjview/), then mapped to the cortical surface using FreeSurfer, and visualized with the PySurfer package (https://pysurfer.github.io/).

Contrasts of interest

We conducted a systematic ALE analysis to evaluate cross-modal recruitment in deaf participants. First, we examined cross-modal recruitment by identifying clusters that were more active in deaf versus hearing participants and identified the auditory areas in hearing participants across all the studies. Next, we categorized studies into two task types to quantify the relative extent of each principle: (i) linguistic tasks, which probe domain-specific language processing in left-lateralized language areas (notably the left STG), and (ii) non-linguistic tasks (such as moving dots and faces), which were analyzed in conjunction with linguistic tasks to identify regions showing domain-general recruitment. We then conducted separate ALE meta-analyses for linguistic and non-linguistic tasks to localize their cross-modal recruitment respectively. Specifically, we performed pairwise ALE contrasts to isolate task-unique activations, and conjunction analyses to identify regions common to both domains. We also calculated a laterality index for each cluster to compare hemispheric asymmetries between tasks. Finally, we performed five control analyses to address potential confounds related to: (i) dynamic properties of linguistic stimuli, (ii) discrepancies between task demands and stimulus types, (iii) variability in the onset of hearing loss, (iv) the impact of a small number of outlier studies, and (v) the influence of publication bias. Detailed information on the studies included in each analysis is provided in Table S2.

Cross-modal recruitment in auditory-deprived cortex

To identify brain regions that consistently show altered activation in deaf compared to hearing participants, we conducted two separate ALE meta-analyses. The first analysis included forty-five studies reporting greater activation in the deaf participants (deaf > hearing), and the second analysis included twenty-three studies reporting reduced activation in the deaf participants (deaf < hearing). The descriptive information of all the studies included in these analyses is shown in Table 1.

To determine whether clusters with altered activation in deaf participants correspond to auditory cortex in hearing participants, we performed the third ALE analysis using five studies involving auditory tasks in hearing participants. In addition, we incorporated functional masks of the auditory cortex derived from term-based online meta-analyses on Neurosynth (https://neurosynth.org/analyses/terms/primary%20auditory/). We combined the results from the third ALE analysis and the auditory masks into a joint auditory mask, and then calculated the percentage of overlapping voxels between the “deaf > hearing” ALE results and the joint auditory mask.

Linguistic tasks versus non-linguistic tasks

To quantify the relative contributions of functional change and functional preservation in deaf participants, we examined the cross-modal recruitment in linguistic and non-linguistic tasks. All the studies were categorized into two groups: tasks with linguistic stimulus (n = 28) and tasks with non-linguistic stimulus (n = 22). Five studies that included both types of stimuli were assigned to both categories. The descriptive information of all the studies included in these analyses was shown in Table 1.

We then conducted four ALE analyses to assess specific cross-modal activations within linguistic tasks and shared cross-modal activations across all tasks. The fourth and fifth ALE analyses localized clusters showing greater activation in deaf compared to hearing participants during linguistic tasks (fourth) and non-linguistic tasks (fifth) respectively (i.e., linguistic: deaf > hearing; non-linguistic: deaf > hearing). The sixth ALE analysis involved pairwise contrasts to identify task-unique activations (linguistic _{deaf > hearing} > non-linguistic _{deaf > hearing} and vice versa). The seventh analysis was a conjunction analysis to identify regions of overlapping cross-modal recruitment across tasks (linguistic _{deaf > hearing} ∩ non-linguistic _{deaf > hearing}). To account for potential differences in sample size across studies, the contrast analysis used 10,000 permutations by randomly reassigning foci between the two datasets. This generated a null distribution, which was used to compute voxel-wise p-values. These values were then converted into z-scores for easier interpretation of significance. Thresholds of p < 0.001 at voxel-level and p < 0.05 at cluster-level inference were applied for multiple comparisons correction.

Lateralization analysis

To compare hemispheric lateralization in cross-modal recruitment, we computed the average laterality index (AveLI), a weighted measure designed to provide a robust estimate of hemispheric dominance during linguistic and non-linguistic tasks. AveLI includes all voxels within a volume of interest (VOI), weighting each by its statistical strength in a fully data-driven manner. AveLI values range from +1 (fully left-lateralized) to −1 (fully right-lateralized).¹⁰⁶

We first defined linguistic and non-linguistic volumes of interest (VOIs) based on the results of the fourth and fifth ALE meta-analyses, respectively. To assess the statistical significance of hemispheric lateralization, we conducted a permutation test in which ALE values from all non-zero voxels within each VOI were randomly reassigned across hemispheres. This procedure was repeated 1,000 times to generate a null distribution of AveLI scores. The AveLI was compared to the null distribution by computing a Z score, which was then used to calculate a p-value, indicating whether the observed lateralization significantly deviated from chance.

Control analysis

To evaluate the robustness of results, we conducted five control analyses: the first addressing potential confounds related to movements-related confounds in the linguistic stimuli, the second accounting for the discrepancies between stimulus types and task demands (e.g., non-linguistic task with linguistic stimulus), the third examining the potential effect of variability in hearing loss onset, the fourth evaluating the influence of outlier experiments on results, and the fifth assessing the impact of publication bias on results.

Control for stimulus motion confounds

Linguistic tasks frequently involve dynamic stimuli (e.g., sign language, lipreading), whereas non-linguistic tasks often use static images. To determine whether the observed differences between tasks could be confounded by motion processing, we subdivided both the linguistic and non-linguistic datasets based on stimulus motion properties. Specifically, dynamic tasks were categorized into dynamic linguistic (n = 20) and dynamic non-linguistic (n = 11) tasks. It should be noted that some subsets contained fewer experiments than the recommended minimum number of experiments (n = 17) for robust ALE cluster detection.⁹⁶

Control for stimulus-task dissociation

While our primary classification distinguished linguistic and non-linguistic studies based on type of stimuli (linguistic or non-linguistic stimuli), some studies used non-linguistic stimuli (e.g., object pictures) with language-related tasks (e.g., Phoneme categorization), and vice versa.^52,107,108 To account for discrepancies between stimuli type and task demands, we classified each study according to task requirements (linguistic: n = 32, non-linguistic: n = 15) regardless of stimulus types. We then repeated the ALE and laterality index analyses.

Control for variability in onset of hearing loss

To control for variability in the onset of hearing loss, we restricted our analysis to studies that explicitly identified prelingual deafness, defined as hearing loss onset before three years of age. Given that prelingual and postlingual deaf individuals are known to exhibit distinct brain activation patterns due to different neurodevelopmental trajectories, this restriction helped mitigate potential confounds. We then separately repeated the ALE and laterality index analyses for the linguistic (n = 23) and non-linguistic (n = 15) datasets.

Robustness test of results

To assess whether the observed clusters were driven by any single influential experiment,⁶⁹ we performed a jackknife sensitivity analysis (also known as leave-one-out analysis). In this procedure, the ALE analysis was iteratively recomputed N times (where N is the number of included experiments), each time omitting one experiment from the dataset. The jackknife score for each cluster was calculated as: (number of iterations in which the cluster remained significant /total number of experiments) ×100%.⁶⁹ This analysis allowed us to quantify the stability of each cluster across the full dataset.

Publication bias assessment of results

To assess the potential impact of publication bias on our results, we conducted fail-safe N (FSN) analyses. In this approach, null studies with results similar to the original studies were progressively added to the dataset until all clusters lost statistical significance.¹⁰⁹ The maximum number of null studies added was set to five times the number of original studies. This process was repeated ten times to calculate a confidence interval.⁶⁹ The FSN score for each cluster was calculated as: (number of null experiments that need to be added until the cluster became insignificant /total number of experiments) ×100%. An FSN score greater than 30% suggests that the results are robust and not significantly influenced by unpublished null studies.¹¹⁰