Non-suicidal self-injury (NSSI), characterized as deliberate, self-inflicted harm without suicidal intent (e.g., cutting, burning, or hitting oneself), and suicide attempts (SA), which involve potentially lethal self-harm with explicit intent to die, represent critical public health challenges globally [1–3]. Recent epidemiological studies indicate that the past-year prevalence of these conditions has reached approximately 18% for non-suicidal self-injury and 10% for suicide attempts among adolescent populations [1, 4, 5]. Longitudinal evidence further establishes NSSI as a strong predictor of subsequent SA, especially in individuals diagnosed with mood disorders or those exhibiting repetitive self-harm patterns [2, 6, 7]. Shared neurocognitive vulnerabilities such as emotion dysregulation and trauma exposure might contribute to this progression, with accumulating evidence suggesting a frequency-dependent relationship where escalating NSSI engagement increases suicide risk [2, 6, 8]. Despite these advances, persistent challenges hinder the mechanistic understanding of current neurobiological models that inadequately explain the transition from NSSI to SA, heterogeneous methodologies for defining self-injury phenotypes compromise cross-study comparability, and insufficient integration of multimodal neuroimaging data limits the identification of transdiagnostic neural substrates underlying these behaviors [9–11]. Addressing these gaps is essential for developing targeted interventions to prevent the progression to fatal suicide.

Neuroimaging advances, particularly multimodal magnetic resonance imaging (MRI) approaches, have elucidated distinct neuroimaging manifestations underlying NSSI and SA. Functional MRI (fMRI) studies have indicated that NSSI involves dual neurofunctional deficits. Affected individuals display heightened activation of the superior temporal gyrus in response to negative emotional stimuli alongside reduced activation of the supplementary motor area during efforts to regulate emotion. This pattern correlates directly with the severity of emotion dysregulation and may promote maladaptive coping behaviors [12]. This aberrant amygdala-anterior cingulate cortex (ACC) connectivity further compounds affective control impairments. Resting-state fMRI (rs-fMRI) corroborates NSSI-specific alterations within visual and default mode networks, offering neuroimaging markers for behavioral risk stratification in adolescent depression [13, 14]. Structurally, diffusion tensor imaging (DTI) has confirmed that reduced fractional anisotropy (FA) across limbic-cortical tracts is critical for emotion regulation, prominently affecting the uncinate fasciculus and cingulum while extending to the cognitive control and sensorimotor pathways [13, 15]. These white matter disruptions parallel gray matter pathology, as evidenced by sMRI, with reduced insular and anterior cingulate cortical thickness alongside diminished right inferior frontal gyrus volume. Crucially, microstructural degradation in frontolimbic white matter tracts predicts attentional impulsivity and chronicity of self-injury, indicating that convergent gray-white matter disorganization fundamentally compromises emotion-behavior integration in individuals with NSSI [13, 16, 17]. SA has a similar convergent pathophysiology but distinct fronto-limbic-striatal circuit dysfunction [18–20]. Gray matter volume (GMV) reductions localize to prefrontal and orbitofrontal cortices with anterior cingulate atrophy, whereas functional alterations manifest as default mode and salience network dysregulation [18–20]. Structural disorganization also involves white matter degradation, which affects prefrontal projections, the internal capsule, and the corpus callosum (CC) [21]. Collectively, these multimodal deficits disrupt both top-down cognitive control and bottom-up emotional signaling, ultimately impairing the regulation of approach-avoidance behaviors across the entire spectrum of self-harm.

Although considerable progress has been made in characterizing the neuroimaging alterations linked to non-suicidal self-injury and suicide attempts, substantial heterogeneity remains across individual studies in terms of the spatial location and effect sizes of the reported neural abnormalities. This methodological variability complicates the replication of results and hinders the establishment of a consensus on transdiagnostic biomarkers. To address this issue, coordinate-based meta-analysis (CBMA) has emerged as a powerful technique for synthesizing neuroimaging findings across disparate studies. This method enhances statistical power by pooling spatial coordinates while simultaneously accounting for inter-study heterogeneity [22, 23]. A notable advancement within this paradigm is the seed-based d mapping (SDM) framework, which enables the quantitative integration of multimodal neuroimaging data. This technique facilitates robust voxel-wise mapping of both structural and functional alterations by employing effect-size weighting and spatial regularization [24–26]. Evidence synthesized through CBMA has begun to clarify consistent patterns of neural dysfunction. In the case of non-suicidal self-injury, a meta-analysis using activation likelihood estimation that integrated nine functional magnetic resonance imaging studies with 359 participants identified significant hyperactivation in two core clusters. These clusters were located in the right medial frontal gyrus, which extends into the rostral anterior cingulate cortex, and the left inferior frontal gyrus, which includes the insular region [27]. As these areas are central to emotional regulation and reward processing, their dysfunction is implicated in core symptoms of non-suicidal self-injury, including impaired emotion control and deficits in reward processing. This distinct functional signature holds promise as a potential biomarker for diagnosis and intervention development [27]. Conversely, in populations with major depressive disorder (MDD) exhibiting suicidality, a recent multimodal meta-analysis combining structural and resting-state fMRI data revealed a coupled structural-functional pathology that distinguished these patients from those with non-suicidal MDD [28]. Specifically, MDD patients with suicidality presented increased GMV alongside functional hypoactivity in the left postcentral gyrus, which may reflect somatosensory integration deficits. They also exhibited decreased gray matter volume concurrent with hypoactivity in the right inferior parietal gyri, suggesting a reduced capacity for cognitive control. Most notably, the right insula in these patients has a reduced gray matter volume yet paradoxically heightened spontaneous activity, indicating a likely disruption in interoceptive awareness mechanisms [28]. These neural alterations are directly linked to clinical presentations in which insular hyperactivity is correlated with the severity of emotional dysregulation, whereas parietal hypoactivity is associated with underlying impulsivity and compromised reward modulation. The right insula appears to be a critical neural hub where structural atrophy and functional maladaptation converge, thereby bridging neuropathological processes across suicidal behaviors [28]. Collectively, these meta-analytic findings indicate that while both NSSI and suicidal MDD patients exhibit dysregulation in key salience network nodes such as the insula and anterior cingulate cortex, their patterns of cortical engagement differ significantly.

Although existing coordinate-based meta-analyses offer valuable insights into the neural signatures of NSSI and SA within specific diagnostic categories, several critical limitations hinder the generalizability of their findings. Existing meta-analytic investigations tend to adopt a narrow focus; some examine only functional alterations in non-suicidal self-injury while omitting structural correlates, whereas others confine their analysis to suicidality present in MDD. This approach overlooks the essential direct comparison of transdiagnostic neural mechanisms that may operate across both NSSI and SA. Moreover, a significant gap remains as no prior CBMA has simultaneously modeled multimodal neuroimaging markers encompassing gray matter morphology, white matter integrity, and functional network dynamics. The absence of such an integrated analysis prevents the identification of both shared and distinct neural pathways that underlie these clinically overlapping but behaviorally divergent phenotypes. Our study aims to address this gap by implementing an SDM framework that systematically compares NSSI and SA across disorders, integrating fMRI, sMRI, and DTI evidence to delineate syndrome-specific alterations versus transdiagnostic vulnerability factors operating across the spectrum of self-harm behaviors. We anticipate that this novel approach will successfully elucidate both convergent and divergent neural circuits across non-suicidal self-injury and suicide attempts. Ultimately, we expect these findings to advance the discovery of biomarkers necessary for developing precision interventions that target and mitigate the progression of self-harm behaviors.

This CBMA was prospectively registered with PROSPERO (ID: CRD420250652566), ensuring transparent methodology and analytical plan disclosure prior to data synthesis. This study implemented systematic review procedures in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) standards to guarantee methodological rigor across the literature search, screening, and data extraction phases [29–31]. A systematic literature search was conducted to identify relevant multimodal neuroimaging studies examining neural alterations associated with non-suicidal self-injury and suicide attempts. Ethical approval was not needed, as the study utilized published data.

Following standardized protocols and established neuroimaging meta-analysis guidelines [32], two independent investigators (C.Z. and R.J.) performed the literature screening, study quality appraisal, and data extraction. The following key study characteristics were documented: authorship, sample size, participant demographics (age/sex distribution), clinical subgroup classifications (NSSI, SA, or comorbid presentations), and diagnostic criteria. The imaging parameters included acquisition protocols, analytical pipelines, statistical thresholds, and peak coordinates alongside their corresponding effect sizes, specifically t values or z scores. All the extracted coordinates and effect metrics were cross-verified to guarantee their accuracy for subsequent SDM analyses.

Statistical analyses were performed using the SDM program v5.15 [25, 33], which implements a standardized processing pipeline. After literature screening, peak coordinates and their corresponding t values from whole-brain comparisons between individuals engaging in self-harm and controls were systematically extracted. These coordinates were then reconstructed within Montreal Neurological Institute standard space via an anisotropic kernel, a process that assigns an effect size to each voxel to best approximate the original findings of each included study. A random-effects mean analysis was employed to combine the study-specific effect size maps into a single meta-analytic mean map. Statistical thresholds were stringently optimized to balance sensitivity and specificity. Following recommendations from the developers of the SDM, we applied thresholds that were stricter than the software defaults. These included a voxel-level threshold of p < 0.001, a peak height threshold of Z ≥ 1.00, and a cluster extent threshold of 10 or more contiguous voxels, with spatial smoothing applied using a 20-mm full width at the half-maximum Gaussian kernel. This conservative analytical approach maintained an estimated false-positive rate of 2.5% while enhancing the power to detect transdiagnostic neural alterations across different self-harm phenotypes.

The reproducibility of the main findings was evaluated using an internal jackknife sensitivity analysis. This method iteratively excludes each individual study and recalculates the statistical maps using the remaining dataset. A neural alteration was deemed robust and reproducible only if it remained statistically significant in every single of these leave-one-out iterations. This stringent criterion ensured that the reported brain regions represented consistent alterations across the spectrum of self-harm behaviors, independent of the influence from any particular study, thereby minimizing the potential impact of study-specific artifacts [33, 34].

To examine the potential modulating effects of demographic and clinical variables, including age, sex, and symptom severity, on neural structure and function, we performed meta-regression analyses utilizing Monte Carlo permutation tests. These analyses were specifically confined to those brain regions that showed significant alterations in the primary meta-analysis, with a focus on neurobiologically relevant circuits. In line with recommendations from the developers of the seed-based d mapping software, a stringent statistical threshold of p < 0.0005 was applied to optimize the balance between sensitivity and robustness while effectively controlling the false discovery rate [33]. This method allowed for a precise characterization of how demographic factors modulate transdiagnostic neural signatures in populations that engage in self-harm.

This multimodal meta-analysis integrates structural and functional neuroimaging evidence across diagnostic boundaries to characterize neural alterations associated with NSSI and SA. We identified several neural alterations that transcended traditional diagnostic categories, including hyperactivity in the PCG, hypoactivity in the mSFG, GMV reductions in the lingual gyrus, and compromised white matter integrity in the CC, BA 48, and amygdala. Advancing age predicted progressive reductions in both functional activity within the mSFG and white matter integrity across frontolimbic tracts. Furthermore, stratification by behavioral phenotype and developmental stage revealed distinct neural profiles. Our findings establish convergent neuroimaging signatures across self-harm presentations while delineating phenotype-specific neural substrates.

The identified hyperactivity in the PCG and hypoactivity in the mSFG represent robust transdiagnostic neural markers across self-harm populations. Elevated intrinsic activity in the PCG directly corresponds to maladaptive self-referential processing, where heightened engagement with internally generated negative cognitions amplifies emotional distress [77–79]. This neural signature impedes effective disengagement from ruminative cycles, as evidenced by aberrant default mode network (DMN) connectivity patterns in pooled self-harm cohorts [19, 80]. Conversely, reduced activity in the mSFG signifies impaired top-down governance over limbic regions [81–83], which likely disrupts conflict monitoring and the suppression of emotional interference during rest [81, 84, 85]. Such executive control deficits may permit the unchecked propagation of distress signals through amygdala-anterior cingulate pathways, effectively lowering the threshold for behavioral disinhibition [86, 87]. This functional imbalance between overactive self-reflection areas and underactive control regions may critically heighten susceptibility to impulsive behaviors [88, 89]. These consistent functional alterations confirm that abnormalities within circuits governing self-focused thinking and behavioral regulation represent fundamental risk factors for self-harm across diagnostic groups [13, 19, 90].

Our meta-analysis revealed a significant reduction in GMV localized specifically to the right lingual gyrus across pooled self-harm cohorts. The lingual gyrus plays a pivotal role in visual processing and emotional stimulus integration [91, 92], particularly in translating affectively salient visual cues into physiological responses [93, 94]. Structural compromise within this region could disrupt the rapid detection of threats and the contextual evaluation of emotionally charged information, thereby potentially contributing to the generation of maladaptive behavioral responses [95, 96]. Evidence indicates that altered lingual gyrus functions and structures are correlated with impaired fear conditioning and altered arousal modulation [97, 98]. Importantly, the present structural findings are consistent with functional neuroimaging studies that demonstrated aberrant engagement of the visual association cortex during emotional provocation tasks [99–101]. This aberrant engagement may be directly associated with the execution of self-injury behaviors [101]. The consistent identification of lingual gyrus vulnerability across traditional diagnostic boundaries suggests that its structural alteration represents a transdiagnostic neural substrate underlying the dysregulation of emotion and behavior in self-harm phenotypes.

Reduced FA in the genu of the CC, BA 48, and right amygdala was identified as a transdiagnostic white matter signature across self-harm populations. The CC serves as the primary commissural tract that integrates interhemispheric communication between the prefrontal and limbic regions [102]. The observed abnormalities in BA 48 encompass several interconnected pathways, including the anterior commissure, uncinate fasciculus, inferior fronto-occipital fasciculus, and insular connections. These pathways collectively support emotion-cognition integration, and their compromise may disrupt this critical function [103–105]. Furthermore, the amygdala serves as a core hub for threat detection and affective arousal [106, 107]. Microstructural compromise in these regions may disrupt top-down cortical regulation of emotional responses and exacerbate bottom-up salience signaling, thereby facilitating maladaptive behavioral disinhibition [108–110]. More specifically, the observed reduction in the integrity of the CC correlates with impaired information transfer between hemispheres during emotion regulation tasks, as evidenced by delayed inhibitory control in response to negative stimuli [108–110]. microstructural degradation within the amygdala is associated with heightened physiological arousal in response to threat cues, a mechanism that may amplify impulsive action tendencies [111, 112]. These alterations converge with functional neuroimaging evidence of fronto-limbic dysregulation, suggesting that white matter disorganization may underlie the inefficient network integration required for the modulation of emotion and behavior [113–115]. The co-occurrence of CC and amygdala abnormalities further indicates that disrupted large-scale structural connectivity rather than isolated regional deficits might mediate vulnerability to behavioral escalation across the self-harm spectrum [13, 19, 116]. Importantly, however, the inclusion of only five diffusion tensor imaging datasets in this analysis necessitates a cautious interpretation of these white matter integrity alterations.

The observed age-dependent reductions in functional activity within the right mSFG, along with declining microstructural integrity in the CC and amygdala, carry critical implications for the design of early intervention strategies. These alterations demonstrate a pattern of progressive deterioration across the lifespan in self-harm populations, suggesting that neurobiological vulnerability escalates when intervention is delayed. The convergence of functional and structural decline within prefrontal-limbic circuits highlights clinically actionable targets for neuromodulation [117, 118]. The mSFG, which exhibited the most pronounced age-related functional decline, represents a promising target for transcranial magnetic stimulation (TMS) protocols to increase top-down regulatory capacity, particularly in adolescent populations [119–121]. Similarly, the microstructural integrity of the CC could serve as a predictive biomarker for stratifying the longitudinal risk of self-harm, thereby enabling a more efficient allocation of clinical resources to high-risk subgroups. Our findings further support the use of age-adapted therapeutic approaches. Interventions for younger individuals should prioritize neuroplasticity-based techniques such as cognitive remediation to harness developmental windows of opportunity. In contrast, treatment protocols for adults may require combinatorial strategies that address accumulated network degradation. Importantly, the transdiagnostic nature of these age-related alterations supports the implementation of standardized screening for frontolimbic integrity across various psychiatric disorders. This could help identify individuals who might be candidates for preemptive neuromodulation before behavioral escalation occurs.

Subgroup analyses revealed distinct neuroimaging signatures that differentiate NSSI from SA phenotypes. In the NSSI cohort, increased functional activity within the left inferior longitudinal fasciculus and median cingulate/paracingulate gyri suggested a heightened integration of visually presented emotionally salient stimuli within limbic circuits. This mechanism may facilitate repetitive self-injury through the amplification of affective resonance [122–124]. This pattern was complemented by reduced activity in the right middle temporal gyrus and mSFG, indicating potential impairments in contextual interpretation and diminished behavioral inhibition. In comparison, cohorts with suicide attempts exhibited hyperactivity in the precuneus alongside gray matter reductions in the right lingual gyrus and the left middle frontal gyrus, reflecting disruptions in bodily self-awareness and compromised cognitive control [125–127]. These phenotype-specific patterns yield actionable clinical insights. The neural signatures associated with NSSI suggest that real-time neuromodulation targets, such as the cingulate cortex, could be engaged through transcranial magnetic stimulation protocols to attenuate emotional hyper-reactivity. Conversely, the biomarkers associated with SA indicate that assessments of structural integrity in the middle frontal gyrus could help guide pharmacological strategies aimed at enhancing executive function. Moreover, developmental-stage subgroup analyses provide an essential context for primary transdiagnostic findings by revealing distinct neurobiological profiles across different age groups. The adult subgroup demonstrated notably stronger convergence than the results from the main pooled analysis did, particularly regarding the consistency of gray matter reductions in the right lingual gyrus and functional alterations in parietal regions. This pattern suggests that data from adult cohorts may contribute more substantially to the effect sizes observed in the primary CBMA, potentially reflecting more established or pronounced neural alterations in adult populations with self-harm. The differential alterations specific to adolescents, particularly within frontal and temporal regions, underscore the critical importance of considering developmental stage in the design and interpretation of future neuroimaging studies on self-harm behaviors. Importantly, the limited number of DTI datasets available for these subgroup analyses highlights an urgent need for future studies to incorporate multimodal imaging to fully characterize neurodevelopmental trajectories.

Taken together, this study establishes three significant conceptual advances in self-harm neuroimaging research. First, our transdiagnostic framework synthesizes evidence across psychiatric disorders to identify neural alterations independent of diagnostic boundaries, addressing critical limitations in disorder-specific meta-analyses. Second, the concurrent integration of functional, structural, and white matter integrity metrics through SDM represents the first multimodal synthesis of neuroimaging evidence for both NSSI and SA, revealing convergent frontolimbic-thalamic dysregulation inaccessible to unimodal approaches. Third, our comprehensive stratification integrated behavioral phenotype and developmental stage subgroup analyses with meta-regression. This approach extends beyond phenotypic comparisons to delineate adolescent-adult neurodevelopmental alterations while identifying accelerated neuroimaging aging trajectories through covariate modeling. These design innovations, combined with the largest pooled neuroimaging cohort to date (46 datasets; N = 3,906), provide unprecedented statistical power to detect transdiagnostic biomarkers while controlling for clinical heterogeneity through rigorous meta-regression.

Several limitations should be considered when interpreting the findings of this study. First, significant heterogeneity existed across the included studies regarding neuroimaging acquisition parameters and analytical pipelines. Differences in MRI field strengths, preprocessing strategies, and statistical correction methods may have influenced the consistency of the observed effects, despite the spatial normalization procedures applied in our meta-analysis. Second, the cross-sectional nature of the available evidence precludes any causal inferences regarding whether the identified neural alterations represent predisposing risk factors or consequences of self-harm behaviors. Future longitudinal research designs are necessary to clarify these neurodevelopmental trajectories. Third, although the SDM provides an optimal method for synthesizing coordinate-based data, its reliance on published peak coordinates might overlook potentially important subthreshold effects. This limitation underscores the value and importance of sharing full statistical maps in open science repositories for future meta-analytic research.

In conclusion, this multimodal meta-analysis provides a novel transdiagnostic framework for understanding the neurobiological underpinnings of self-harm behaviors. By integrating evidence across psychiatric diagnoses and neuroimaging modalities, we demonstrate that both shared and distinct neural circuit dysregulations underlie non-suicidal self-injury and suicide attempts. These findings significantly advance the field beyond disorder-specific models and offer a more nuanced neurobiological account of self-harm pathophysiology. Crucially, our identification of phenotype-specific neural signatures and age-related progressive alterations provides a critical foundation for developing targeted interventions and personalized prevention strategies. This work not only bridges a fundamental gap between neuroimaging research and clinical practice but also establishes a new direction for biomarker discovery in suicide prevention, ultimately contributing to more effective and timely interventions for individuals at risk.