Development and validation of a prediction model for non-suicidal self-injury in adolescents based on functional near-infrared spectroscopy and neuropsychological assessment

Abstract

BACKGROUND

Non-suicidal self-injury (NSSI) is prevalent among adolescents worldwide, yet current identification methods rely heavily on subjective self-report tools that are prone to concealment and fail to capture the neurobiological underpinnings of the behavior. Integrating objective neuroimaging markers with neuropsychological assessment may improve early identification and risk stratification of NSSI.

AIM

To develop and validate a prediction model for NSSI behavior in adolescents based on functional near-infrared spectroscopy (fNIRS) and neuropsychological assessment indicators.

METHODS

A retrospective study was conducted, including 312 adolescents (156 NSSI cases and 156 controls) who visited the psychology department of a tertiary hospital from March 2021 to March 2024. All participants completed fNIRS assessment (verbal fluency task) and neuropsychological evaluation [Difficulties in Emotion Regulation Scale (DERS), Childhood Trauma Questionnaire (CTQ), Barratt Impulsiveness Scale-11 (BIS-11), Adolescent Self-Rating Life Events Check List (ASLEC)]. Univariate analysis was used to screen variables, and multivariate logistic regression was employed to establish the prediction model. A nomogram was constructed, and internal validation was performed using the Bootstrap method. The model’s performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curve, and decision curve analysis (DCA).

RESULTS

The NSSI group showed significantly lower prefrontal oxyhemoglobin concentration changes and activation integral values compared to the control group (P < 0.001). Multivariate logistic regression revealed that left dorsolateral prefrontal cortex activation integral value [odds ratio (OR) = 0.72, 95%CI: 0.58-0.89], DERS non-acceptance of emotional responses dimension (OR = 1.15, 95%CI: 1.08-1.23), CTQ emotional neglect dimension (OR = 1.12, 95%CI: 1.05-1.19), BIS-11 motor impulsiveness dimension (OR = 1.18, 95%CI: 1.09-1.28), and ASLEC interpersonal relationship dimension (OR = 1.09, 95%CI: 1.03-1.16) were independent predictors of NSSI. The prediction model based on these factors achieved an AUC of 0.891 (95%CI: 0.854-0.928), with sensitivity of 82.7% and specificity of 81.4%. Bootstrap internal validation showed a corrected AUC of 0.876. The calibration curve demonstrated good consistency between predicted and actual probabilities, and DCA indicated favorable clinical net benefit.

CONCLUSION

The prediction model based on fNIRS prefrontal activation indicators and neuropsychological assessment demonstrates good predictive performance for adolescent NSSI and can provide objective evidence for early identification and risk stratification of NSSI.

Key Words: Non-suicidal self-injury; Adolescent; Functional near-infrared spectroscopy; Prediction model; Nomogram; Prefrontal cortex

Core Tip: Non-suicidal self-injury (NSSI) in adolescents is difficult to identify early using self-report measures alone. This study developed and internally validated a prediction model integrating functional near-infrared spectroscopy prefrontal activation indicators with multidimensional neuropsychological assessments. Reduced left dorsolateral prefrontal cortex activation, emotional non-acceptance, childhood emotional neglect, motor impulsivity, and interpersonal stress were identified as independent predictors of NSSI. The nomogram-based model demonstrated good discrimination, calibration, and clinical utility, providing an objective and practical tool for early risk identification and stratified intervention in adolescent NSSI.

INTRODUCTION

Non-suicidal self-injury (NSSI) refers to the intentional and direct damage to one’s own body tissue without suicidal intent[1]. Globally, the lifetime prevalence of NSSI among adolescents is approximately 17%-18%, with detection rates in Chinese adolescents ranging from 22%-27%, showing an increasing trend in recent years[2,3]. NSSI not only causes physical harm but is also an important predictor of suicidal ideation and behavior. Adolescents with a history of NSSI have a 3-4 times increased risk of future suicide attempts[4]. Therefore, early identification of high-risk individuals for NSSI is of great significance for suicide prevention and promoting adolescent mental health.

Currently, NSSI identification primarily relies on clinical interviews and self-report scales, which have certain subjectivity and limitations. Some adolescents may conceal self-injury behaviors due to shame or fear of blame, leading to missed diagnoses[5]. Additionally, traditional assessment methods have difficulty capturing the neurobiological basis of NSSI, limiting deeper understanding of its mechanisms. Therefore, exploring objective and quantifiable biological markers for NSSI identification and prediction is of significant value.

Functional near-infrared spectroscopy (fNIRS) is an emerging non-invasive brain functional imaging technology that reflects regional brain activation by measuring changes in cortical oxyhemoglobin (HbO₂) and deoxyhemoglobin concentrations[6]. Compared to functional magnetic resonance imaging (fMRI), fNIRS has advantages such as low cost, good portability, insensitivity to motion artifacts, and the ability to detect in natural environments, making it particularly suitable for adolescent research[7]. Previous studies have shown that NSSI patients exhibit prefrontal cortex dysfunction, particularly reduced activation of the dorsolateral prefrontal cortex (DLPFC) in emotion regulation and impulse control[8].

From a psychological perspective, emotion regulation difficulties, childhood trauma, impulsivity, and negative life events are important risk factors for NSSI[9-12]. Emotion regulation difficulties are considered a core feature of NSSI, with approximately 90% of NSSI episodes related to negative emotion regulation[13]. Childhood trauma can lead to abnormal neurodevelopment and emotion regulation deficits, increasing NSSI risk[14]. Highly impulsive individuals are more likely to engage in impulsive self-injury during emotional distress[15]. Negative life events, as stressors, can trigger NSSI episodes[16].

These psychological risk factors are not independent but interact within a unified neurobiological framework. We propose the prefrontal-limbic imbalance model as the overarching theoretical scaffold of the present study: Reduced DLPFC activation impairs top-down inhibitory control over the amygdala, thereby lowering the threshold at which negative emotional states escalate toward behavioral dysregulation. Within this framework, emotional non-acceptance [Difficulties in Emotion Regulation Scale (DERS)] and motor impulsivity [Barratt Impulsiveness Scale-11 (BIS-11)] represent the psychological and behavioral expressions of this regulatory deficit, respectively. Childhood emotional neglect [Childhood Trauma Questionnaire (CTQ)] functions as a developmental antecedent that disrupts the normal maturation of prefrontal-limbic circuitry, while interpersonal stressors [Adolescent Self-Rating Life Events Check List (ASLEC)] act as proximal environmental triggers that interact with these pre-existing vulnerabilities to precipitate NSSI episodes. This integrative model explains why each domain contributes uniquely to NSSI risk and, critically, why neuroimaging and neuropsychological indicators should be considered together rather than in isolation.

A further rationale for incorporating fNIRS into a prediction model concerns the incremental value it provides over purely psychological assessments. Self-report scales, while validated and clinically informative, are susceptible to response bias and deliberate concealment, concerns that are heightened in adolescents with NSSI who often minimize self-injurious behavior out of shame. fNIRS-derived prefrontal activation indices, by contrast, constitute an objective and involuntary neural signal that cannot be consciously manipulated. Beyond circumventing reporting bias, neuroimaging indicators capture the biological substrate of risk, potentially identifying vulnerable individuals before maladaptive coping patterns become sufficiently entrenched to register reliably on behavioral scales. Prior prediction models relying solely on psychometric instruments have yielded area under the receiver operating characteristic curve (AUC) values in the range of 0.70-0.80; the hypothesis that incorporating an objective neural marker of prefrontal regulatory capacity would provide incremental discriminative value is directly tested in the present study.

However, there is currently a lack of NSSI prediction models that integrate neuroimaging indicators and neuropsychological assessments. Single indicators have limited predictive efficacy, while integrating multidimensional information may improve prediction accuracy. Based on this, this study adopted a retrospective case-control design, integrating fNIRS prefrontal activation indicators and neuropsychological assessment scales to establish a prediction model for adolescent NSSI. The model was visualized through a nomogram, and its predictive performance and clinical application value were evaluated to provide objective evidence for early identification and risk stratification of NSSI.

MATERIALS AND METHODS

Study design and subjects

A retrospective study design was adopted, collecting data from adolescents who visited the outpatient or inpatient psychology department of our hospital from March 2021 to March 2024. NSSI group inclusion criteria: (1) Meeting fifth version of the Diagnostic and statistical manual of mental disorders NSSI research diagnostic criteria; (2) Age 12-18 years; (3) At least 5 NSSI behaviors in the past 12 months; and (4) Completed fNIRS assessment and neuropsychological evaluation. Exclusion criteria: (1) Comorbid schizophrenia, bipolar disorder, or organic mental disorders; (2) Intellectual disability (IQ < 70); (3) Substance abuse or dependence; and (4) Unqualified fNIRS data quality. The control group consisted of adolescents without a history of NSSI who visited during the same period, matched 1:1 with the NSSI group by age and gender. This study was approved by the hospital ethics committee (No. 2025-449-01).

fNIRS data collection and processing

Assessment was performed using a near-infrared brain functional imaging system (Hitachi ETG-4000, 52 channels). Probes were placed according to the international 10-20 system, covering bilateral prefrontal cortex. The experimental task employed the verbal fluency task (VFT): Participants were asked to verbalize as many words as possible belonging to a specified semantic category (such as “animals” or “fruits”) within 60 seconds, alternating between task periods (60 seconds) and rest periods (30 seconds) for a total of 3 task cycles. HbO₂ concentration changes in bilateral DLPFC (CH22-25, CH27-30) and ventrolateral prefrontal cortex (VLPFC, CH11-14, CH38-41) were recorded during task periods.

Data preprocessing included: Bandpass filtering (0.01-0.5 Hz), baseline correction, and motion artifact removal. Specifically, the ETG-4000 system sampled at 10 Hz. The bandpass filter applied was a zero-phase Butterworth filter (0.01-0.5 Hz) to remove physiological noise and low-frequency drift. Baseline was defined as the mean HbO₂ signal during the 10-second pre-task rest window immediately preceding each task epoch. The inter-optode distance was standardized at 3 cm across all source-detector pairs. Motion artifacts were identified using an automated algorithm: Any channel exhibiting an abrupt signal change exceeding 0.2 mmol/L·mm within a single sampling interval (100 ms) was flagged. Flagged segments were subsequently reviewed by a trained research assistant blinded to group allocation. Channels were excluded from analysis if artifacts coincided with task epochs or exceeded 20% of the total recording duration. Participants were excluded entirely if more than 30% of channels within the region of interest (bilateral DLPFC or VLPFC) failed quality criteria, or if the coefficient of variation of the raw signal exceeded 10% in the majority of task-relevant channels. Applying these criteria, 14 participants (8.2% of the initially screened sample) were excluded due to unacceptable fNIRS data quality prior to final group allocation.

Main indicators included: (1) HbO₂ concentration change peak value; (2) Activation integral value (area under the HbO₂ curve during the full 60-second task epoch relative to the corrected baseline); and (3) Activation latency.

Neuropsychological assessment

DERS: Total 36 items covering 6 dimensions including difficulty in emotional awareness, lack of emotional clarity, non-acceptance of emotional responses, limited access to strategies, impulse control difficulties, and difficulties engaging in goal-directed behaviors. Higher scores indicate more severe emotion regulation difficulties[17].

CTQ: Total 28 items covering 5 dimensions including emotional abuse, physical abuse, sexual abuse, emotional neglect, and physical neglect[18].

BIS-11: Total 30 items covering 3 dimensions including attentional impulsiveness, motor impulsiveness, and non-planning impulsiveness[19].

ASLEC: Total 27 items covering 6 dimensions including interpersonal relationships, academic pressure, punishment, loss, health adaptation, and others[20].

Other data collection

Demographic data (gender, age, years of education, family structure), clinical data (comorbidities, family history of mental illness, previous treatment history), and NSSI characteristics (onset age, duration, self-injury methods, frequency) were collected through the electronic medical record system.

Statistical analysis

Data analysis was performed using SPSS 26.0 and R 4.3.0 software. Continuous data were expressed as mean ± SD or median (P25-P75), with independent samples t-test or Mann-Whitney U test for between-group comparisons; categorical data were expressed as n (%), with χ² test for between-group comparisons. With NSSI occurrence as the dependent variable, variables with P < 0.1 in univariate analysis were included in multivariate logistic regression (forward stepwise method) to screen independent predictors. A prediction model was constructed based on regression results and visualized with a nomogram. Model validation included: (1) Discrimination: Calculating AUC and 95%CI; (2) Calibration: Plotting calibration curves and calculating Hosmer-Lemeshow goodness-of-fit; (3) Clinical utility: Decision curve analysis (DCA); and (4) Internal validation: 1000 Bootstrap resampling iterations. P < 0.05 was considered statistically significant.

RESULTS

Basic characteristics of study subjects

A total of 312 adolescents were included, with 156 cases in the NSSI group and 156 cases in the control group. There were no statistically significant differences between the two groups in gender, age, and years of education (P > 0.05), indicating comparability. In the NSSI group, the mean onset age was (13.2 ± 1.4) years, duration was (22.5 ± 14.8) months, and the main self-injury methods were cutting (68.6%), scratching (15.4%), burning (8.3%), and others (7.7%) (Table 1).

Table 1 Comparison of basic characteristics between two groups, n (%)/mean ± SD.

Item	NSSI group (n = 156)	Control group (n = 156)	t/χ2	P value
Gender			0.103	0.748
Male	42 (26.9)	45 (28.8)
Female	114 (73.1)	111 (71.2)
Age (years)	15.3 ± 1.6	15.1 ± 1.7	1.052	0.294
Years of education (years)	8.6 ± 1.5	8.8 ± 1.4	-1.186	0.237
Only child	82 (52.6)	78 (50.0)	0.205	0.651
Divorced parents	38 (24.4)	25 (16.0)	3.425	0.064
Comorbid depression	98 (62.8)	32 (20.5)	56.82	< 0.001
Comorbid anxiety	72 (46.2)	28 (17.9)	28.56	< 0.001
Family history of mental illness	35 (22.4)	18 (11.5)	6.528	0.011

NSSI: Non-suicidal self-injury.

Comparison of fNIRS indicators between two groups

The NSSI group showed significantly lower HbO₂ concentration change peak values and activation integral values in bilateral DLPFC and VLPFC compared to the control group (P < 0.01), with significantly longer activation latency (P < 0.01). The most significant between-group difference was observed in left DLPFC activation integral value (t = -6.825, P < 0.001) (Table 2).

Table 2 Comparison of functional near-infrared spectroscopy indicators between two groups, mean ± SD.

Indicator	NSSI group (n = 156)	Control group (n = 156)	t value	P value
Left DLPFC
HbO2 peak (μM)	0.042 ± 0.018	0.068 ± 0.022	-10.82	< 0.001
Activation integral (μM/s)	1.85 ± 0.72	3.12 ± 0.86	-13.68	< 0.001
Activation latency (s)	18.6 ± 5.2	12.5 ± 4.1	11.25	< 0.001
Right DLPFC
HbO2 peak (μM)	0.038 ± 0.016	0.062 ± 0.020	-11.52	< 0.001
Activation integral (μM/s)	1.68 ± 0.65	2.85 ± 0.78	-13.92	< 0.001
Activation latency (s)	19.2 ± 5.5	13.2 ± 4.3	10.56	< 0.001
Left VLPFC
HbO2 peak (μM)	0.035 ± 0.015	0.052 ± 0.018	-8.86	< 0.001
Activation integral (μM/s)	1.52 ± 0.58	2.35 ± 0.68	-11.25	< 0.001
Right VLPFC
HbO2 peak (μM)	0.032 ± 0.014	0.048 ± 0.017	-8.72	< 0.001
Activation integral (μM/s)	1.42 ± 0.55	2.18 ± 0.62	-11.08	< 0.001

DLPFC: Dorsolateral prefrontal cortex; VLPFC: Ventrolateral prefrontal cortex; NSSI: Non-suicidal self-injury; HbO₂: Cortical oxyhemoglobin.

Comparison of neuropsychological assessment results between two groups

The NSSI group showed significantly higher total scores and dimensional scores on DERS, CTQ, BIS-11, and ASLEC compared to the control group (P < 0.01) (Table 3).

Table 3 Comparison of neuropsychological assessment results between two groups, mean ± SD.

Scale and dimension	NSSI group (n = 156)	Control group (n = 156)	t value	P value
DERS total score	112.5 ± 18.6	72.8 ± 15.2	20.25	< 0.001
Difficulty in emotional awareness	16.8 ± 4.2	12.5 ± 3.5	9.62	< 0.001
Lack of emotional clarity	14.2 ± 3.8	10.2 ± 3.2	9.86	< 0.001
Non-acceptance of emotional responses	18.5 ± 4.5	11.8 ± 3.6	14.25	< 0.001
Limited access to strategies	22.6 ± 5.2	14.5 ± 4.2	14.86	< 0.001
Impulse control difficulties	18.2 ± 4.8	11.2 ± 3.5	14.38	< 0.001
Difficulties engaging in goal-directed behaviors	17.5 ± 4.2	12.6 ± 3.8	10.52	< 0.001
CTQ total score	52.8 ± 14.5	35.6 ± 10.2	11.86	< 0.001
Emotional abuse	10.5 ± 3.8	7.2 ± 2.5	8.86	< 0.001
Physical abuse	7.8 ± 3.2	5.8 ± 1.8	6.62	< 0.001
Emotional neglect	15.6 ± 4.5	10.2 ± 3.2	11.86	< 0.001
Physical neglect	10.2 ± 3.5	7.5 ± 2.6	7.52	< 0.001
BIS-11 total score	75.2 ± 11.5	58.6 ± 9.8	13.52	< 0.001
Attentional impulsiveness	18.5 ± 4.2	14.2 ± 3.5	9.62	< 0.001
Motor impulsiveness	24.6 ± 5.5	18.2 ± 4.2	11.25	< 0.001
Non-planning impulsiveness	28.5 ± 5.8	22.5 ± 4.8	9.72	< 0.001
ASLEC total score	48.6 ± 15.2	28.5 ± 10.5	13.25	< 0.001
Interpersonal relationships	12.8 ± 4.5	7.2 ± 3.2	12.35	< 0.001
Academic pressure	11.5 ± 3.8	6.8 ± 2.5	12.42	< 0.001

NSSI: Non-suicidal self-injury; DERS: Difficulties in Emotion Regulation Scale; CTQ: Childhood Trauma Questionnaire; BIS-11: Barratt Impulsiveness Scale-11; ASLEC: Adolescent Self-Rating Life Events Checklist.

Univariate logistic regression analysis of NSSI influencing factors

Univariate logistic regression analysis showed that divorced parents, comorbid depression, comorbid anxiety, family history of mental illness, left DLPFC activation integral value, right DLPFC activation integral value, all dimensions of DERS, all dimensions of CTQ, all dimensions of BIS-11, and all dimensions of ASLEC were associated with NSSI occurrence (P < 0.1) (Table 4).

Table 4 Univariate logistic regression analysis of non-suicidal self-injury influencing factors.

Variable	B	SE	OR (95%CI)	Wald	P value
Divorced parents	0.525	0.286	1.69 (0.97-2.96)	3.37	0.066
Comorbid depression	1.856	0.252	6.40 (3.90-10.49)	54.25	< 0.001
Comorbid anxiety	1.352	0.248	3.87 (2.38-6.29)	29.72	< 0.001
Family history of mental illness	0.825	0.325	2.28 (1.21-4.32)	6.45	0.011
Left DLPFC activation integral value	-1.125	0.162	0.32 (0.24-0.44)	48.25	< 0.001
Right DLPFC activation integral value	-1.052	0.158	0.35 (0.26-0.47)	44.36	< 0.001
DERS non-acceptance of emotional responses	0.185	0.025	1.20 (1.15-1.26)	54.76	< 0.001
DERS limited access to strategies	0.152	0.022	1.16 (1.12-1.21)	47.72	< 0.001
CTQ emotional neglect	0.142	0.022	1.15 (1.10-1.20)	41.68	< 0.001
CTQ emotional abuse	0.125	0.028	1.13 (1.07-1.20)	19.94	< 0.001
BIS-11 motor impulsiveness	0.162	0.025	1.18 (1.12-1.23)	42.02	< 0.001
BIS-11 attentional impulsiveness	0.135	0.028	1.14 (1.08-1.21)	23.25	< 0.001
ASLEC interpersonal relationships	0.152	0.028	1.16 (1.10-1.23)	29.47	< 0.001
ASLEC academic pressure	0.098	0.022	1.10 (1.06-1.15)	19.85	< 0.001

DLPFC: Dorsolateral prefrontal cortex; DERS: Difficulties in Emotion Regulation Scale; CTQ: Childhood Trauma Questionnaire; BIS-11: Barratt Impulsiveness Scale-11; ASLEC: Adolescent Self-Rating Life Events Checklist; OR: Odds ratio.

Multivariate logistic regression analysis of NSSI influencing factors

Variables with P < 0.1 in univariate analysis were included in the multivariate logistic regression model (forward stepwise method). The results showed that left DLPFC activation integral value [odds ratio (OR) = 0.72], DERS non-acceptance of emotional responses dimension (OR = 1.15), CTQ emotional neglect dimension (OR = 1.12), BIS-11 motor impulsiveness dimension (OR = 1.18), and ASLEC interpersonal relationship dimension (OR = 1.09) were independent predictors of NSSI (P < 0.05) (Table 5).

Table 5 Multivariate logistic regression analysis of non-suicidal self-injury influencing factors.

Variable	B	SE	OR (95%CI)	Wald	P value
Left DLPFC activation integral value	-0.328	0.108	0.72 (0.58-0.89)	9.22	0.002
DERS non-acceptance of emotional responses	0.142	0.032	1.15 (1.08-1.23)	19.69	< 0.001
CTQ emotional neglect	0.112	0.035	1.12 (1.05-1.19)	10.24	0.001
BIS-11 motor impulsiveness	0.168	0.042	1.18 (1.09-1.28)	16.00	< 0.001
ASLEC interpersonal relationships	0.086	0.032	1.09 (1.03-1.16)	7.23	0.007

Model χ² = 186.52, P < 0.001; Hosmer-Lemeshow test χ² = 8.256, P = 0.409; Nagelkerke R² = 0.586. DLPFC: Dorsolateral prefrontal cortex; DERS: Difficulties in Emotion Regulation Scale; CTQ: Childhood Trauma Questionnaire; BIS-11: Barratt Impulsiveness Scale-11; ASLEC: Adolescent Self-Rating Life Events Checklist; OR: Odds ratio.

Construction of prediction model and nomogram

Based on the multivariate logistic regression results, a prediction model for NSSI was constructed with the regression equation: Logit (P) = -0.328 × left DLPFC activation integral value + 0.142 × DERS non - acceptance of emotional responses + 0.112 × CTQ emotional neglect + 0.168 × BIS-11 motor impulsiveness + 0.086 × ASLEC interpersonal relationships - 2.856. A nomogram was drawn (Figure 1) to intuitively calculate individual NSSI occurrence risk.

Figure 1 Nomogram for non-suicidal self-injury prediction model. This nomogram was constructed based on a multivariate logistic regression model to predict the risk of non-suicidal self-injury. The nomogram includes five predictor variables: Left dorsolateral prefrontal cortex activation integral value, Difficulties in Emotion Regulation Scale non-acceptance of emotional responses score, Childhood Trauma Questionnaire emotional neglect score, Barratt Impulsiveness Scale-11 motor impulsiveness score, and Adolescent Self-Rating Life Events Checklist interpersonal relationships score. DLPFC: Dorsolateral prefrontal cortex; DERS: Difficulties in Emotion Regulation Scale; CTQ: Childhood Trauma Questionnaire; BIS-11: Barratt Impulsiveness Scale-11; ASLEC: Adolescent Self-Rating Life Events Checklist.

Validation of prediction model

Discrimination: Receiver operating characteristic curve analysis showed that the prediction model’s AUC was 0.891 (95%CI: 0.854-0.928), with an optimal cutoff value of 0.485 at maximum Youden index, sensitivity of 82.7%, and specificity of 81.4% (Figure 2A).

Figure 2 Validation of non-suicidal self-injury prediction model. A: Receiver operating characteristic curve: Area under the curve (AUC) = 0.891 (95%CI: 0.854-0.928); optimal cutoff = 0.485 with sensitivity 82.7% and specificity 81.4%; B: Calibration curve: Hosmer-Lemeshow test χ² = 8.256, P = 0.409, indicating good calibration. Bootstrap validation (1000 iterations): Corrected AUC = 0.876 (95%CI: 0.835-0.917); C: Decision curve: The model provides net benefit over treat-all or treat-none strategies within threshold probability range 015-0.85, demonstrating clinical utility. ROC: Receiver operating characteristic; NSSI: Non-suicidal self-injury.

Calibration: Hosmer-Lemeshow goodness-of-fit test χ² = 8.256, P = 0.409, indicating good model calibration. The calibration curve showed good consistency between predicted and actual probabilities (Figure 2B).

Internal validation: Bootstrap resampling 1000 times yielded a corrected AUC of 0.876 (95%CI: 0.835-0.917), indicating good model stability.

Clinical utility: DCA showed that when the threshold probability ranged from 0.15-0.85, using this model for NSSI risk prediction provided net benefit superior to “treat all” or “treat none” strategies (Figure 2C).

Neuropsychological correlation analysis: To test the hypothesized relationship between prefrontal function and behavioral impulsivity, Pearson correlation analysis was conducted between left DLPFC activation integral values and BIS-11 motor impulsiveness scores. A significant negative correlation was observed across the full sample (r = -0.38, P < 0.001), indicating that lower prefrontal activation was associated with higher motor impulsivity. This correlation was stronger within the NSSI group (r = -0.42, P < 0.001) than within the control group (r = -0.21, P = 0.009), suggesting that the coupling between frontal regulatory capacity and impulsive behavior is particularly pronounced among individuals engaging in self-injury.

DISCUSSION

This study integrated fNIRS prefrontal activation indicators and multidimensional neuropsychological assessments to establish a prediction model for adolescent NSSI. The results showed that left DLPFC activation integral value, DERS non-acceptance of emotional responses dimension, CTQ emotional neglect dimension, BIS-11 motor impulsiveness dimension, and ASLEC interpersonal relationship dimension were independent predictors of NSSI. The prediction model based on these factors achieved an AUC of 0.891, demonstrating good discrimination and calibration. Bootstrap internal validation confirmed model stability and reliability, and DCA confirmed its clinical utility value. This study is the first to combine fNIRS objective indicators with neuropsychological assessments for NSSI prediction, providing a new tool for early identification and risk stratification of NSSI.

The fNIRS results showed that the NSSI group had significantly reduced prefrontal cortex activation, particularly in the left DLPFC. The DLPFC is a key brain region for executive function, cognitive control, and emotion regulation, playing an important role in top-down regulation of limbic system activity[21]. Previous fMRI studies have also found that NSSI patients show reduced DLPFC activation during emotion regulation tasks and abnormal prefrontal-amygdala functional connectivity[22]. This study employed the VFT, which requires participation of working memory, semantic retrieval, and executive control, effectively activating the prefrontal cortex[23]. The reduced prefrontal activation in the NSSI group during this task reflects deficits in their executive function and cognitive control abilities. Notably, the left DLPFC activation integral value was the only brain functional indicator as an independent predictor in the multivariate model, suggesting that this indicator may be a potential biological marker for NSSI.

From a neurodevelopmental perspective, the prefrontal cortex continues to develop rapidly during adolescence, with myelination processes not yet complete, making executive function and emotion regulation abilities of this age group relatively weak[24]. Meanwhile, the limbic system (particularly the amygdala) is relatively mature during adolescence, leading to a state of developmental imbalance characterized by enhanced emotional reactivity with insufficient regulatory capacity[25]. Adolescents with NSSI may exhibit more pronounced developmental imbalance, with prefrontal function deficits making it difficult for them to effectively regulate negative emotions, increasing the risk of impulsive self-injury behavior. fNIRS, as an objective and convenient detection method, can be used to assess prefrontal functional status, providing neuroimaging evidence for NSSI risk assessment.

Among neuropsychological assessment indicators, the DERS non-acceptance of emotional responses dimension entered the final prediction model. Non-acceptance of emotional responses refers to individuals’ denial, resistance, and rejection of their own negative emotional experiences, manifested in cognitions such as “I shouldn’t feel this way” or “this emotion is wrong”[26]. Research shows that emotional non-acceptance can lead to amplification and persistence of emotional experiences, forming a vicious cycle and increasing the likelihood of adopting maladaptive coping strategies (such as NSSI)[27]. In contrast, individuals who can accept and be aware of their emotions are more likely to adopt adaptive emotion regulation strategies. This finding suggests that interventions targeting emotional acceptance (such as mindfulness training and acceptance and commitment therapy) may be important targets for NSSI prevention and treatment.

The CTQ emotional neglect dimension was another important predictor. Emotional neglect refers to caregivers’ failure to provide adequate emotional support, attention, and responsiveness, representing a covert form of childhood trauma[28]. Compared with overt traumas such as emotional abuse and physical abuse, emotional neglect is often more difficult to identify, but its impact on psychological development is equally profound. Research shows that emotional neglect can lead to lack of secure attachment, impeded development of emotion regulation abilities, and reduced self-worth, increasing NSSI risk[29]. From a neurobiological perspective, early emotional neglect can affect the normal development of the prefrontal-limbic system, leading to functional abnormalities in emotion regulation-related neural circuits[30]. This study’s results emphasize the importance of paying attention to childhood emotional neglect history in NSSI risk assessment.

The BIS-11 motor impulsiveness dimension reflects individuals’ impulsivity at the behavioral level, characterized by lack of thinking before acting and difficulty inhibiting immediate responses[31]. Highly motor-impulsive individuals, when experiencing strong negative emotions, are more likely to adopt immediate, unreflective behavioral responses, including NSSI[32]. Motor impulsiveness is closely related to inhibitory control function of the prefrontal cortex, and this study also found reduced prefrontal activation in the NSSI group, with the correlation between the two supporting the neuropsychological model of NSSI. In clinical practice, interventions targeting impulsivity (such as impulse control training and delay of gratification training) may help reduce NSSI occurrence.

The ASLEC interpersonal relationship dimension reflects stress events adolescents experience in peer relationships, teacher-student relationships, and other aspects[33]. Adolescence is a critical period for social relationship development, and the quality of peer relationships has important impacts on mental health. Interpersonal relationship stress (such as peer rejection, friendship breakup, and bullying) can trigger intense negative emotional experiences, and individuals with insufficient emotion regulation abilities may adopt NSSI as a coping strategy[34]. Additionally, interpersonal relationship stress may also reinforce the social function of NSSI (such as gaining attention and avoiding social demands), maintaining the cycle of self-injury behavior[35]. This finding suggests that paying attention to adolescents’ interpersonal relationship status and providing social support is of great significance for NSSI prevention.

The prediction model constructed in this study integrated information from both neuroimaging and neuropsychological levels, achieving an AUC of 0.891, significantly higher than previous prediction models based on single-dimensional assessment (typically AUC between 0.70-0.80)[36]. The model’s high discrimination and calibration were supported by Bootstrap internal validation, and DCA also confirmed its clinical utility. The visual presentation of the nomogram makes the model convenient for clinical application, allowing clinicians to quickly calculate NSSI risk probability based on patients’ indicator scores. This model can be used for early identification of NSSI, high-risk population screening, and risk stratification, guiding rational allocation of prevention and intervention resources.

An important consideration concerns the specificity of the observed prefrontal hypoactivation to NSSI vs a general psychopathology burden. Given that 62.8% of NSSI participants had comorbid depression and 46.2% had comorbid anxiety, it is plausible that reduced DLPFC activation partly reflects these comorbidities rather than NSSI per se. To examine this, we conducted a sensitivity analysis in which comorbid depression and comorbid anxiety were added as covariates to the multivariate logistic regression. The left DLPFC activation integral value remained a significant independent predictor (OR = 0.76, 95%CI: 0.61-0.95, P = 0.013), suggesting that its association with NSSI is not entirely attributable to comorbid affective disorders, although the modest attenuation of effect size indicates partial overlap. While prefrontal hypoactivation is reported in both depression and anxiety, the specific combination of reduced DLPFC activation during an executive task with motor impulsivity and emotional non-acceptance may characterize a neuropsychological profile conferring particular vulnerability to the behavioral act of self-injury. We explicitly acknowledge, however, that a definitive test of NSSI-specificity would require a three-arm design including a psychiatric comparison group without NSSI history, which constitutes a priority for future research.

From a clinical application perspective, this study provides a multidimensional NSSI risk assessment framework. First, fNIRS assessment can provide objective prefrontal function evaluation, compensating for the subjective limitations of self-report scales; second, multidimensional neuropsychological assessment can comprehensively capture psychological risk factors for NSSI; finally, the prediction model and nomogram provide quantitative tools for comprehensive risk assessment. In clinical practice, it is recommended to routinely conduct NSSI risk screening for adolescent patients, perform detailed neuropsychological assessment and necessary fNIRS testing for high-risk individuals, conduct risk stratification based on the prediction model, and develop individualized prevention and intervention programs.

This study has the following limitations: (1) Most critically, the study lacks an external validation cohort. Despite the use of Bootstrap resampling to correct for optimism bias, internal validation cannot substitute for validation in an independent sample from a different clinical setting or population. The corrected AUC of 0.876 should therefore be regarded as an upper-bound estimate of generalizable performance rather than a confirmed benchmark. The single-center origin of our sample from a tertiary psychiatric facility further introduces potential selection bias toward more severe or treatment-seeking cases, which may inflate apparent discriminative ability relative to community or primary-care settings. We strongly caution against premature clinical implementation until prospective multicenter external validation is completed; (2) The retrospective case-control study design has inherent limitations and cannot determine causal relationships; (3) fNIRS can only detect cortical surface activity and cannot evaluate functions of deep brain structures (such as the amygdala); and (4) The current predictor set, while theoretically grounded and clinically accessible, does not exhaust the NSSI risk landscape. Genetic polymorphisms in serotonin transporter and monoamine oxidase genes, perfectionism as a personality trait linked to emotional distress intolerance, sleep disturbances that bidirectionally interact with negative affect and inhibitory control, and peer NSSI exposure given the well-documented social contagion dynamics of self-injury in adolescent networks all represent plausible and measurable candidates for inclusion in future model iterations.

CONCLUSION

This study established a prediction model for adolescent NSSI based on fNIRS prefrontal activation indicators and neuropsychological assessment. Left DLPFC activation integral value, DERS non-acceptance of emotional responses, CTQ emotional neglect, BIS-11 motor impulsiveness, and ASLEC interpersonal relationships are independent predictors of NSSI. The model demonstrates good discrimination (AUC = 0.891), calibration, and clinical utility, and can provide objective evidence for early identification, risk stratification, and individualized intervention of adolescent NSSI.