Deep learning approach to predict developmental outcomes of non-suicidal self-injury: An ERP study

Deep learning approach to predict developmental outcomes of non-suicidal self-injury: An ERP study | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Deep learning approach to predict developmental outcomes of non-suicidal self-injury: An ERP study Fei Yin, Feng Si, Wenlong Jiang, Shuhui Huo, Binquan Wang, Li Liu, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5784879/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Identifying predictors of developmental outcomes in non-suicidal self-injury (NSSI) is crucial and goes beyond tracking its progression. EEG technology is notable for its consistent and objective neurophysiological recordings in NSSI detection. Using ERP components in deep learning models for predicting these outcomes is still underexplored. Methods Twenty-six in the remission group (RG), twenty-nine in the aggravation group (AG), and twenty-seven in the healthy group (HG) completed the affective Stroop task with EEG. N2 and P3 component differences were analyzed across groups, and the EEGNet model was used to assess NSSI developmental outcomes. Result A significant interaction was observed between group and emotion on N2 (F (2, 79) = 16.934, p < 0.001, η 2 = 0.300). Under neutral stimuli, N2 was smallest in HG, larger in RG, and largest in AG, while for negative stimuli, N2 in HG was smaller than in RG and AG. A significant group effect on P3 was noted (F (2, 79) = 7.607, p < 0.001, η 2 = 0.161), with HG exhibiting larger P3 compared to RG and AG. The N2 under neutral stimuli achieved the highest classification accuracy (94.31%). Conclusion The findings indicate that NSSI is linked to cognitive processing deficits, including impaired control and resource allocation to stimuli. Additionally, N2 amplitudes were shown to reliably predict developmental outcomes in NSSI. Non-suicidal Self-injury EEG Developmental outcomes EEGNet ERP Figures Figure 1 Figure 2 Figure 3 Figure 4 Background Non-suicidal self-injury (NSSI) refers to deliberate and direct damage to body tissue without suicidal intent [ 1 ]. While not intended to result in death, evidence suggests that NSSI can predict future suicide attempts and fatalities [ 2 , 3 ]. Lifetime prevalence rates of NSSI among Chinese adolescents have been reported to range from 16–29% [ 4 , 5 ]. The classification of NSSI as an independent category in the DSM-5 underscores its significance and the necessity for further research. NSSI typically begins in early adolescence and progresses into diverse developmental trajectories over time. Group-Based Trajectory Models have identified distinct patterns, such as high-risk and low-risk groups [ 6 ]. Another study involving 13- to 19-year-olds identified three patterns of NSSI frequency: Low Stable, Moderate Fluctuating, and High Persistent [ 7 ]. Similarly, a year-long study of 3,381 Chinese adolescents found two trajectory groups: a faller group (87.4%) and a higher group (12.7%) [ 8 ]. These longitudinal studies highlight the varied developmental pathways of NSSI, emphasizing the importance of tailored prevention and intervention strategies. While identifying trajectories is crucial, determining predictors of developmental outcomes holds even greater importance. Current forecasting methods largely rely on validated scales. From the perspectives of family system theory and cumulative risk, family conflict risk has been identified as a key predictor of NSSI addiction [ 6 ]. Other studies highlight the roles of interpersonal factors, such as unstable relationships and parental criticism, as predictors of high NSSI fluctuations [ 8 ]. Personal factors, including impulsive behaviors, depression [ 8 ], emotion regulation difficulties, self-compassion deficits [ 9 ], psychiatric symptoms, and comorbidities [ 7 ], also contribute significantly. However, reliance on subjective self-reports introduces bias and reduces diagnostic accuracy. Objective measurements based on individual characteristics offer a more reliable alternative, presenting a promising avenue for predicting NSSI outcomes with greater precision. Behavioral findings from various cognitive tasks have provided insights into cognitive control in subgroups of individuals with NSSI. Frequent NSSI has been associated with poorer performance on the Criticism Gambling Task, suggesting that impulsive decision-making in response to criticism may serve as a critical indicator of NSSI progression [ 10 ]. Additionally, high-severity NSSI groups have been found to exhibit working memory deficits, whereas low-severity groups demonstrate impaired inhibitory control [ 11 ]. Electroencephalogram (EEG) devices allow for the collection of stimulus-evoked EEGs, or event-related potentials (ERPs), through specifically designed experimental tasks. Previous studies have reported that adolescents with NSSI, compared to HC, exhibit increased N2 amplitudes [ 12 ] and decreased P3 amplitudes [ 13 ], indicating significant cognitive control dysfunction. Further, Zhao et al. identified that adolescents with NSSI displayed smaller N2 and larger P3 responses during negative emotional face stimuli [ 14 ]. However, no ERP-based studies currently address developmental outcomes of NSSI, highlighting the potential of investigating neurocognitive factors using ERP methodologies. ERPs offer high temporal resolution, making them valuable for studying cognitive processes with millisecond precision [ 15 , 16 ]. The N2 component, located in the anterior central brain region, is indicative of conflict processing, with its amplitude reflecting the degree of cognitive control [ 17 ]. Cognitive processing involves the activation and coordination of multiple brain regions. For instance, the affective Stroop task studied via fMRI has demonstrated activation in regions such as the amygdala, insula, and prefrontal cortex, alongside cognitive networks involving the prefrontal cortex, superior temporal lobe, and insula [ 18 ]. This study seeks to examine the cognitive processing features of NSSI outcome groups using an affective Stroop paradigm, which is a reliable tool for assessing cognitive functions [ 19 ]. Given that negative emotions can impair cognitive performance [ 20 , 21 ], the affective Stroop task is a classical paradigm used in recent years [ 22 , 23 ]. Washburn et al. noted that individuals frequently engaging in NSSI might have a behavioral addiction [ 24 ], sharing neurobiological roots with addiction[ 25 ]. Drug addiction is marked by cue reactivity [ 26 ], shown through behavioral and neurobiological methods like ERP. A prior study indicated that adolescents with NSSI might show altered behavioral and brain responses to NSSI cues [ 27 ]. Therefore, a novel affective Stroop paradigm which contains neutral and NSSI clues images as stimuli was adapted to evaluate the effects of emotional valence on cognitive tasks across NSSI outcome groups. The primary objective was to analyze the dynamic neural mechanisms in these groups through ERP analysis. Deep learning has demonstrated significant capability in automatically learning and extracting features from raw data, making it highly applicable to EEG signal analysis [ 28 , 29 ]. Compared to conventional methods, deep learning techniques offer several advantages, including robust feature extraction and representation, end-to-end training, and superior generalization. Among these, EEGNet, a convolutional neural network (CNN) specifically designed for Brain-Computer Interface (BCI) applications [ 30 ], has shown exceptional performance in processing task-related EEG datasets. These include applications such as P3 visual-evoked potentials [ 31 ], movement-related cortical potentials (MRCP) [ 32 ], and sensory motor rhythms (SMR) [ 33 ]. In this study, ERP components such as N2 and P3 were extracted from the affective Stroop dataset for comparative analysis using EEGNet. The secondary objective was to predict developmental outcomes of NSSI by employing these ERP components as input features to the EEGNet model. Methods Participants Participants were selected from an ongoing longitudinal study designed to predict developmental outcomes of NSSI. This study tracked the trajectories of NSSI in 1960 freshmen (mean age = 18.13 ± 1.65) across three waves, each separated by a six-month interval. NSSI was evaluated based on DSM-5 criteria [ 34 ] for the past six months, which includes eight specific behaviors: cutting/stabbing, hair pulling, head banging, biting, cauterizing, scratching, hitting, and rubbing until bleeding occurred. Latent class analysis revealed three subgroups: healthy group (HG) (n = 1627, 83.01%), remission group (RG) (n = 276, 14.08%), and aggravation group (AG) (n = 57, 2.90%). Ninety freshmen were recruited from the longitudinal study, with 30 participants in each subgroup (HG, RG, AG). Eight participants were excluded: two for accuracy below 75%, three for z-scores above 3.29, and three due to excessive EEG artifacts in over 25% of trials. The final sample included 82 participants (mean age 18.17 ± 1.73; 76 females): 27 in HG, 26 in RG, and 29 in AG. Eligibility criteria included normal or corrected vision, no color blindness, no history of alcohol or drug abuse, and no mental illness. Those who met these criteria underwent structured clinical interviews based on DSM-5 criteria conducted by counselors for group assignment confirmation. Participants volunteered for the study, provided informed consent, could withdraw at any time, and were compensated with cash. Ethical approval was obtained from the Ethics Committee of Harbin Medical University Daqing Campus, and the study adhered to the ethical standards outlined in the World Medical Association Declaration of Helsinki. Task paradigm All participants were required to complete the affective Stroop task during EEG recording. The task was created using E-prime 3.0 (see Fig. 1 ). It consisted of two blocks, each containing 120 neutral trials and 120 negative trials. The trials were presented in a pseudo-randomized order within each block. The emotional stimuli included 30 neutral pictures selected from the Chinese Affective Picture System (CAPS) [ 35 ] and 30 neutral pictures sourced from NSSI images of the participants. The two sets of emotional pictures differed significantly in valence [F (1, 58) = 136.51, P < 0.001] but were similar in arousal [F (1, 58) = 5.17, P = 0.21]. The literal stimuli consisted of single words presented in one of four colors (red, yellow, blue, or green). Two experimental conditions were used: the congruent condition, where each color word was displayed in its corresponding color, and the incongruent condition, where each color word was presented in a color other than the one it referred to (e.g., the word "red" displayed in yellow). Each trial began with a white “+” displayed for 500–700 ms in the center of the screen on a grey background, followed by an emotional picture presented for 1000 ms, then a blank screen for 600 ms, followed by a color word displayed for 200 ms, and a final blank screen for 1500 ms. Participants were required to determine whether the word and color were congruent. If they were congruent, participants pressed the "D" key with their left index finger; if they were incongruent, participants pressed the "J" key with their right index finger. Both response time and accuracy were recorded in this study. EEG recording and processing The electroencephalogram (EEG) data were recorded using a Neuroscan ESI 64-channel system, equipped with an Ag/AgCl electrode cap and the extended international 10–20 electrode placement. The system was configured with a bandpass filter of 0.05–100 Hz and a sampling rate of 500 Hz. A left mastoid electrode was used as the online reference, which was later re-referenced offline to the common average [ 36 ]. Vertical electrooculogram (EOG) was recorded from electrodes placed above and below the left eye, while horizontal EOG was captured from electrodes positioned at the outer canthi. Electrode impedance was kept below 5 kΩ throughout the task. Experimental conditions, equipment, and physiological responses can introduce interference that affects the collected signals. To minimize these effects, preprocessing steps were applied to the EEG data. EOG blink artifacts were corrected using linear regression estimation [ 37 ]. The EEG signals were band-pass filtered between 0.05 Hz and 30 Hz and segmented into 1000 ms epochs, with a 200 ms pre-stimulus baseline. Epochs with voltage deviations exceeding ± 80 µV were excluded from the ERP analysis. Separate averages for the congruent and incongruent conditions were computed for neutral and negative stimuli, respectively. Only trials with correct responses were included in the calculation of average ERPs. ERP analysis Based on ERP components linked to cognition as reported in previous research [ 38 – 40 ], as well as the characteristics of the ERP waveforms observed in the current study, specific ERP components were selected for further analysis. The amplitudes of the N2 (260–320 ms; FZ, FCZ, PZ, CPZ, PZ) and P3 (320–370 ms; F3, F1, FZ, F2, F4, FC3, FC1, FCZ, FC2, FC4) components were examined. For each ERP component, the mean amplitude within the specified time window was calculated. Statistical analysis A repeated-measures analysis of variance (ANOVA) was conducted to assess accuracy, reaction time (RT), and ERP component amplitude across different subgroups. The analysis included a three-way repeated-measures design with group (healthy, declining, and worsening) as the between-subject factor, and emotion (neutral, negative) and congruency (congruent, incongruent) as within-subject factors. Statistical analyses were performed using repeated-measures ANOVAs with Greenhouse-Geisser corrections at a significance level of 0.05. Significant interaction effects were followed by simple effect analysis, with pairwise comparisons adjusted using the Bonferroni method and an alpha value of 0.05 divided by the number of comparisons. Deep learning algorithm In the model, a 64-channel EEG signal was used as the input, and the group served as the output. The model was trained and tested through a deep convolutional layer followed by a deep separable convolutional layer. This network extracts intrinsic characteristic information from the waveform using convolutional kernels, enabling the prediction of developmental outcomes associated with NSSI. The datasets were split in two ways: randomly into three subsets—60% for training, 20% for validation, and 20% for testing—ensuring that no participant's data appeared in more than one subset, and employing 5-fold cross-validation. Results Behavioral performance Table 1 presents the accuracy and reaction times (RTs) for the different groups. Table 1 The accuracies and average reaction times in the different groups (M ± SD) Variable Group Neutral stimuli Congruent Incongruent Negative stimuli Congruent Incongruent Accuracies HG 0.94 ± 0.06 0.84 ± 0.05 0.94 ± 0.05 0.81 ± 0.04 RG 0.94 ± 0.07 0.84 ± 0.09 0.94 ± 0.11 0.79 ± 0.02 AG 0.95 ± 0.05 0.84 ± 0.06 0.94 ± 0.07 0.80 ± 0.06 Reaction time (ms) HG 609 ± 91 646 ± 101 599 ± 113 624 ± 106 RG 623 ± 111 659 ± 109 606 ± 112 610 ± 122 AG 666 ± 91 712 ± 97 681 ± 147 714 ± 141 The ANOVA of accuracy revealed a significant main effect of emotion type (F (1,79) = 21.494, p < 0.001, η 2 = 0.214), indicating that accuracy was higher for neutral stimuli compared to negative stimuli (p < 0.001). The main effect of congruency was also significant (F (1, 79) = 682.573, p < 0.001, η 2 = 0.898), showing that accuracy was higher for the congruent condition than for the incongruent condition (p < 0.001) (Fig. 2 A). The ANOVA of RTs showed a significant main effect of group (F (2,79) = 4.145, p < 0.05, η 2 = 0.095), with slower RTs observed in the AG compared to the HG (p < 0.05). The main effect of congruency was also significant (F (1, 79) = 65.734, p < 0.001, η 2 = 0.454), indicating that RTs were slower for the incongruent condition compared to the congruent condition (p < 0.001) (Fig. 2 B). ERP results Table 2 presents the average amplitude for each ERP component in the different groups. Table 2 The average amplitude for each ERP component in the different groups (M ± SD) Component Group Neutral stimuli Congruent Incongruent Negative stimuli Congruent Incongruent N2 HG 1.09 ± 0.83 -0.41 ± 0.86 -0.41 ± 0.83 -0.27 ± .81 RG -2.16 ± 0.85 -1.66 ± 0.88 -3.70 ± 0.85 -3.63 ± 0.83 AG -3.84 ± 0.80 -3.98 ± 0.83 -3.30 ± 0.80 -2.89 ± 0.78 P3 HG 4.12 ± 0.75 2.10 ± 0.82 2.07 ± 0.78 1.39 ± 0.76 RG 1.05 ± 0.77 0.19 ± .83 -1.98 ± 0.79 -2.87 ± 0.78 AG -0.25 ± 0.73 -1.20 ± 0.79 -1.43 ± 0.75 -1.44 ± 0.74 N2 amplitude The results, presented in Fig. 3 , revealed a significant main effect of emotion type (F (1, 79) = 7.271, p < 0.01, \(\:\eta\:\) 2 = 0.084), where the N2 elicited by negative stimuli was larger than that elicited by neutral stimuli. The main effect of group was also significant (F (2, 79) = 7.091, p < 0.01, η 2 = 0.152). Specifically, the N2 in the HG was significantly smaller than in the RG and AG (ps < 0.05). The interaction between group and emotion was significant (F (2, 79) = 16.934, p < 0.001, η 2 = 0.300). Under neutral stimuli, the N2 in HG was significantly smaller than in the RG (p < 0.05) and AG (p < 0.01), with the RG also being significantly smaller than the AG (p < 0.05). Under negative stimuli, the N2 in HG was significantly smaller than in the RG (p < 0.01) and AG (p < 0.05). The interaction between group, emotion, and congruency was significant (F (2, 79) = 4.068, p < 0.05, η 2 = 0.093). Under neutral stimuli, the N2 in HG was smaller than in AG for both congruent (p < 0.001) and incongruent (p < 0.01) conditions, and the N2 in HG was also smaller than in RG for the congruent condition (p < 0.01). Under negative stimuli, the N2 in HG was smaller than in RG for both congruent (p < 0.01) and incongruent (p < 0.01) conditions. P3 amplitude As shown in Fig. 4 , the main effect of emotion type was significant (F (1, 79) = 73.858, p < 0.001, η 2 = 0.483), with the P3 elicited by neutral stimuli being larger than that elicited by negative stimuli. The main effect of group was also significant (F (2, 79) = 7.607, p < 0.001, η 2 = 0.161), where the P3 in HG was significantly larger than in RG (p < 0.01) and AG (p < 0.01). The main effect of congruency was significant (F (1, 79) = 15.61, p < 0.001, η 2 = 0.165), with the P3 elicited by the congruent condition being significantly larger than that elicited by the incongruent condition (p 0.05, η 2 = 0.034). Simple effects analysis revealed that, under neutral stimuli, the P3 elicited by the congruent condition in HG was larger than in AG (p < 0.001). Under negative stimuli, the P3 elicited by the congruent condition in HG was larger than in RG (p < 0.01) and AG (p < 0.05), while the P3 elicited by the incongruent condition in HG was larger than in RG (p < 0.001). Prediction results on the EEGNet We identified four ERP components during neutral and negative stimulation using the EEGNet model. Table 2 presents the classification accuracies for N2, P3, and LPP with respect to anxiety under the two affective stimuli. Under neutral stimuli, the ERP components achieved peak accuracies of 94.31% for N2 and 89.60% for P3, while under negative stimuli, the accuracies were 73.16% for N2 and 62.57% for P3. These results indicate that N2 performs best under neutral stimuli, with all components showing reduced accuracy under negative stimuli. Discussions This study explored cognitive function in college students with varying NSSI developmental outcomes using an affective Stroop task, supporting the role of cognitive function as a predictor of NSSI outcomes. Specifically, the AG exhibited slower reaction times and larger N2 and smaller P3 amplitudes for neutral stimuli compared to RG and HG. The N2 elicited by neutral stimuli, which is associated with identifying developmental outcomes in NSSI, demonstrated high reliability. Our research utilized neutral and negative images with distinct valence differences as stimuli in the affective Stroop task. The results indicated that negative stimuli impaired cognitive performance relative to neutral stimuli, leading to slower RTs and lower accuracy. This suggests that negative emotions should be closely monitored in individuals with NSSI to prevent adverse outcomes related to impaired cognitive response [ 41 ]. The AG exhibited slower RTs compared to the HG, suggesting lower processing efficiency and impaired cognitive control. Participants in the AG required more time to perform conflict detection and task-related activities. No significant differences in accuracy were observed between groups, indicating that NSSI primarily affected processing efficiency rather than task effectiveness, although this finding requires further confirmation due to the small sample size. Negative stimuli evoked greater N2 amplitudes than neutral stimuli, demonstrating that negative stimuli can rapidly and automatically capture attention and utilize attentional resources. Yuan et al. reported that individuals display heightened sensitivity to negative stimuli, processing them at an earlier stage [ 42 ]. The results showed a significant difference in N2 amplitudes among the HG, RG, and AG, with the amplitude exhibiting a gradient change pattern across these groups. These findings indicate that participants in RG and AG, particularly AG, exhibit abnormalities in cognitive control mechanisms. Hanslmayr et al. observed that individuals responded to increased conflict levels by temporarily enhancing cortical responses to task-relevant information, which improved cognitive control performance instead of suppressing task-irrelevant information [ 43 ]. Participants in RG and AG were not required to inhibit or overcome interference caused by affective stimuli. This absence of inhibition enhanced attention to target stimuli, resulting in larger N2 components. In contrast, HG displayed superior cognitive control, allowing effective inhibition of task-irrelevant information and successful interference management during task completion [ 44 ]. Under neutral stimuli, participants in AG exhibited more pronounced N2 amplitude deficits compared to RG, with significant deficits in incongruent N2 amplitude relative to HG. These results suggest that N2 amplitude serves as a sensitive marker for distinguishing HG from NSSI groups, as well as between RG and AG, providing an effective measure for predicting NSSI outcomes. The P3 amplitudes were markedly smaller under negative stimuli or incongruent conditions, aligning with findings from previous studies [ 45 ]. The P3 component is widely utilized as a biomarker in research on cognitive resource allocation associated with emotion and cognition [ 46 ]. It is instrumental in assessing inhibitory processes, such as in the emotional signal stop task, where larger P3 amplitudes signify stronger inhibitory control [ 47 ]. During conflict processing, suppressing responses to sad faces has been shown to reduce P3 amplitude [ 48 ]. In the present study, the decreased P3 amplitude indicates that cognitive resources are heavily consumed by negative stimuli or conflicts during the Stroop task. Additionally, the P3 exhibited a gradient downward trend across the HG, RG, and AG, suggesting a progressive decline in the brain's resource utilization efficiency in the NSSI group [ 49 ]. Under neutral and negative stimuli, AG participants demonstrated more severe P3 amplitude deficits than HG in congruent conditions. Similarly, RG participants displayed significant P3 amplitude deficits under negative stimuli, indicating notable cognitive control impairments compared to HG. These results highlight the importance of continued monitoring for RG participants, even though their NSSI behaviors are alleviated. Notably, AG participants showed an intermediate pattern under negative stimuli, with N2 and P3 amplitudes positioned between those of HG and RG. This may suggest that negative images become desensitized for AG participants due to NSSI behaviors, resulting in reduced vigilance compared to HG [ 50 ]. Alternatively, AG participants may experience more pronounced cognitive dysfunction, requiring additional resources to process negative stimuli, thereby impairing Stroop task performance [ 47 , 51 ]. Although the precise mechanisms remain unclear, these findings provide valuable insights into potential developmental trajectories. In the RG and AG subgroups, a significant difference was observed in N2 amplitude but not in P3 amplitude, suggesting that AG participants experience greater difficulties with cognitive control over stimuli rather than resource allocation. This finding highlights impaired cognitive control as a central factor in the cognitive dysfunction associated with NSSI development, corroborating earlier studies [ 52 , 53 ]. The EEGnet model used in this study applies depthwise separable convolutional layers to optimize EEG feature extraction, encompassing spatial filtering and filter bank construction. Notably, the analysis demonstrated that N2 amplitudes elicited by neutral stimuli achieved the highest classification accuracy (94.31%) for predicting developmental outcomes of NSSI among the three ERP components evaluated. These findings establish N2 amplitudes as reliable indicators for identifying developmental outcomes of NSSI, contributing to early detection and intervention strategies. This study suggests that N2 amplitude could serve as a biomarker for predicting NSSI outcomes, offering a potential avenue for developing interventions to reduce addiction risk. Improving cognitive control by addressing deficits in the N2 component may effectively mitigate the risk of addiction in individuals with NSSI. This study has several limitations. First, the relatively small sample size may influence the reliability and robustness of the findings. Second, the participants were exclusively college students, which may restrict the generalizability of the results to broader populations. Third, the follow-up period for individuals with NSSI was limited to one year, which may be insufficient for comprehensively evaluating developmental outcomes. A longer follow-up period could potentially reveal different developmental trajectories of NSSI. Conclusions This study employed a novel affective Stroop task to examine the neurocognitive mechanisms underlying different NSSI outcomes and to predict these outcomes using ERP components. The findings demonstrated that NSSI is associated with abnormal cognitive processing, characterized by impaired cognitive control and resource allocation. Notably, N2 amplitudes emerged as reliable indicators for identifying developmental outcomes of NSSI. These results provide an objective foundation for identifying NSSI, predicting its developmental progression, and formulating early intervention strategies. Declarations Informed consent was obtained from all participants in the study. This study was approved by the Ethics Committee of Harbin Medical University (HMUDQ202201585). Acknowledgements The authors thank all the students and schools who participated in this study. Author contributions Fei Yin conceived of the study, participated in its design and performed the statistical analysis coordination and drafted the manuscript; Feng Si participated in the design and interpretation of the data; Wenlong Jiang conceived of the study, and participated in its design and helped to draft the manuscript. Shuhui Huo, Binquan Wang, Li Liu and Nan Yang participated in the design and coordination of the study and performed the measurement; Jianqin Cao revised paper and polished language. All authors read and approved the final manuscript. Funding This research was supported by the National Natural Science Foundation of China (Grant No. 72204065) and the Philosophy and Social Science Foundation of Heilongjiang Provincial (Grant No. 21SHC216). Data availability The datasets produced in the course of this study are not publicly accessible in order to protect the privacy of the subjects. However, they can be obtained from the corresponding author upon reasonable request. Consent for publication Not applicable. Competing interests The authors report no conflict of interests. References Nock MK. Self-injury. Annu Rev Clin Psychol. 2010;6:339–63. Ribeiro JD, et al. Self-injurious thoughts and behaviors as risk factors for future suicide ideation, attempts, and death: a meta-analysis of longitudinal studies. Psychol Med. 2016;46(2):225–36. Wolff JC, et al. Emotion dysregulation and non-suicidal self-injury: A systematic review and meta-analysis. Eur Psychiatry. 2019;59:25–36. Liang K, et al. Nonsuicidal self-injury behaviour in a city of China and its association with family environment, media use and psychopathology. Compr Psychiatry. 2022;115:152311. Zhao H, et al. Cyberbullying victimization and nonsuicidal self-injury in adolescents: Testing a moderated mediating model of emotion reactivity and dispositional mindfulness. J Affect Disord. 2022;299:256–63. Li L, Yang H. Heterogeneity in Adolescents' Non-Suicidal Self-Injury Behaviour Trajectories Based on the Group-Based Trajectory Model and a Decision Tree Analysis of Family-Related Determinants. Psychol Res Behav Manag. 2023;16:3359–71. Mason GE, Auerbach RP, Stewart JG. Predicting the trajectory of non-suicidal self-injury among adolescents. J Child Psychol Psychiatry, 2024. Wang B, et al. Developmental Trajectories of Nonsuicidal Self-Injury in Adolescence and Intrapersonal/Interpersonal Risk Factors. J Res Adolesc. 2017;27(2):392–406. Farrell BCT, Ewing L, Hamza CA. Examining trajectories of nonsuicidal self-injury across the first year of university. J Affect Disord. 2024;367:202–9. Allen KJD, et al. Frequency of nonsuicidal self-injury is associated with impulsive decision-making during criticism. Psychiatry Res. 2019;271:68–75. Fikke LT, Melinder A, Landrø NI. Executive functions are impaired in adolescents engaging in non-suicidal self-injury. Psychol Med. 2011;41(3):601–10. Wen Y, et al. Cognitive Impairment in Adolescent Major Depressive Disorder With Nonsuicidal Self-Injury: Evidence Based on Multi-indicator ERPs. Front Hum Neurosci. 2021;15:637407. Liu H, et al. Prefrontal cortex neural activity predicts reduction of non-suicidal self-injury in adolescents with major depressive disorder: An event related potential study. Front Neurosci. 2022;16:972870. Zhao L, et al. Changes in emotion-related EEG components and brain lateralization response to negative emotions in adolescents with nonsuicidal self-injury: An ERP study. Behav Brain Res. 2023;445:114324. Li F, et al. The Time-Varying Networks in P300: A Task-Evoked EEG Study. IEEE Trans Neural Syst Rehabil Eng. 2016;24(7):725–33. Zinchenko A, et al. Emotion and goal-directed behavior: ERP evidence on cognitive and emotional conflict. Soc Cogn Affect Neurosci. 2015;10(11):1577–87. Heidlmayr K, Kihlstedt M, Isel F. A review on the electroencephalography markers of Stroop executive control processes. Brain Cogn. 2020;146:105637. Raschle NM, et al. Investigating the Neural Correlates of Emotion-Cognition Interaction Using an Affective Stroop Task. Front Psychol. 2017;8:1489. Scarpina F, Tagini S. The Stroop Color and Word Test. Front Psychol. 2017;8:557. Uher R, et al. Increasing cognitive load reduces interference from masked appetitive and aversive but not neutral stimuli. PLoS ONE. 2014;9(4):e94417. Chen L, et al. Impact of acute stress on human brain microstructure: An MR diffusion study of earthquake survivors. Hum Brain Mapp. 2013;34(2):367–73. Flaisch T, Steinhauser M, Schupp HT. Adaptive cognitive control attenuates the late positive potential to emotional distractors. NeuroImage. 2019;200:51–8. Dragan WU, et al. Temperament and neural activation during the affective Stroop task: A functional connectivity study. Pers Indiv Differ. 2022;186:111385. Washburn JJ, et al. Measuring the urge to self-injure: preliminary data from a clinical sample. Psychiatry Res. 2010;178(3):540–4. Luijten M, et al. Systematic review of ERP and fMRI studies investigating inhibitory control and error processing in people with substance dependence and behavioural addictions. J Psychiatry Neurosci. 2014;39(3):149–69. Drummond DC. What does cue-reactivity have to offer clinical research? Addiction. 2000;95(Suppl 2):S129–44. Burke TA, et al. Emotional response inhibition to self-harm stimuli interacts with momentary negative affect to predict nonsuicidal self-injury urges. Behav Res Ther. 2021;142:103865. Ho CSH, et al. Interpretable deep learning model for major depressive disorder assessment based on functional near-infrared spectroscopy. Asian J Psychiatr. 2024;92:103901. Tian X, et al. Social anxiety prediction based on ERP features: A deep learning approach. J Affect Disord. 2024;367:545–53. Lawhern VJ, et al. EEGNet: a compact convolutional neural network for EEG-based brain-computer interfaces. J Neural Eng. 2018;15(5):056013. Polich J. Updating P300: an integrative theory of P3a and P3b. Clin Neurophysiol. 2007;118(10):2128–48. Gordon SM, et al. Informed decomposition of electroencephalographic data. J Neurosci Methods. 2015;256:41–55. Lotte F. Signal Processing Approaches to Minimize or Suppress Calibration Time in Oscillatory Activity-Based Brain–Computer Interfaces. Proceedings of the IEEE, 2015. 103(6): pp. 871–890. American Psychiatric Association, K. Diagnostic and statistical manual of mental disorders: DSM-5. Volume 5. American psychiatric association Washington, DC; 2013. Bai L, Huang MH. The Development of Native Chinese Affective Picture System-A pretest in 46 College Students. Chin Mental Health J. 2005;11(19):719–22. Keil A, et al. Committee report: publication guidelines and recommendations for studies using electroencephalography and magnetoencephalography. Psychophysiology. 2014;51(1):1–21. Semlitsch HV, et al. A solution for reliable and valid reduction of ocular artifacts, applied to the P300 ERP. Psychophysiology. 1986;23(6):695–703. Wang X, Shangguan C, Lu J. Time course of emotion effects during emotion-label and emotion-laden word processing. Neurosci Lett. 2019;699:1–7. Folstein JR, Van Petten C. Influence of cognitive control and mismatch on the N2 component of the ERP: a review. Psychophysiology. 2008;45(1):152–70. Baetens K, et al. N400 and LPP in spontaneous trait inferences. Brain Res. 2011;1418:83–92. Hart SJ, et al. Emotional priming effects during Stroop task performance. NeuroImage. 2010;49(3):2662–70. Yuan J, et al. Are we sensitive to valence differences in emotionally negative stimuli? Electrophysiological evidence from an ERP study. Neuropsychologia. 2007;45(12):2764–71. Hanslmayr S, et al. The electrophysiological dynamics of interference during the Stroop task. J Cogn Neurosci. 2008;20(2):215–25. Xinyu G, et al. Weakening of recognition memory caused by task-irrelevant emotional encoding context can be modulated by individuals' inhibitory control. Personality Individual Differences. 2018;134:201–9. Yang K, et al. Extremely negative emotion interferes with cognition: Evidence from ERPs and time-varying brain network. J Neurosci Methods. 2023;396:109922. Vanlessen N, et al. Positive emotion broadens attention focus through decreased position-specific spatial encoding in early visual cortex: evidence from ERPs. Cogn Affect Behav Neurosci. 2013;13(1):60–79. Senderecka M. Emotional enhancement of error detection-The role of perceptual processing and inhibition monitoring in failed auditory stop trials. Cogn Affect Behav Neurosci. 2018;18(1):1–20. Yu F, et al. Decreased response inhibition to sad faces during explicit and implicit tasks in females with depression: Evidence from an event-related potential study. Psychiatry Res Neuroimaging. 2017;259:42–53. Duncan CC, Kosmidis MH, Mirsky AF. Closed head injury-related information processing deficits: an event-related potential analysis. Int J Psychophysiol. 2005;58(2–3):133–57. Baker A, et al. Does habituation matter? Emotional processing theory and exposure therapy for acrophobia. Behav Res Ther. 2010;48(11):1139–43. Ahumada-Méndez F, et al. Affective modulation of cognitive control: A systematic review of EEG studies. Physiol Behav. 2022;249:113743. Liu J, et al. Sensitivity to reward and punishment in adolescents with repetitive non-suicidal self-injury: The role of inhibitory control. Int J Clin Health Psychol. 2024;24(2):100456. Liu RT. Characterizing the course of non-suicidal self-injury: A cognitive neuroscience perspective. Neurosci Biobehav Rev. 2017;80:159–65. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5784879","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":399446398,"identity":"546929f2-2b2f-47f0-85c8-13ecab40aa4f","order_by":0,"name":"Fei Yin","email":"","orcid":"","institution":"Harin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Fei","middleName":"","lastName":"Yin","suffix":""},{"id":399446399,"identity":"0a700217-c5aa-47b8-a6e4-3b1aa27ac336","order_by":1,"name":"Feng Si","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Feng","middleName":"","lastName":"Si","suffix":""},{"id":399446400,"identity":"e8836f07-39f5-46c5-8536-91ad8718e1fe","order_by":2,"name":"Wenlong Jiang","email":"","orcid":"","institution":"The Third People’ Hospital of Daqing","correspondingAuthor":false,"prefix":"","firstName":"Wenlong","middleName":"","lastName":"Jiang","suffix":""},{"id":399446401,"identity":"f0eb0a1a-2bdc-4565-8d3d-26a13c171cd8","order_by":3,"name":"Shuhui Huo","email":"","orcid":"","institution":"Harin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Shuhui","middleName":"","lastName":"Huo","suffix":""},{"id":399446402,"identity":"39dcb99d-ecce-4786-8961-395b7b62b5fb","order_by":4,"name":"Binquan Wang","email":"","orcid":"","institution":"Harin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Binquan","middleName":"","lastName":"Wang","suffix":""},{"id":399446403,"identity":"ba105a72-13bb-4c84-8010-5789a46c3ffb","order_by":5,"name":"Li Liu","email":"","orcid":"","institution":"Harin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Li","middleName":"","lastName":"Liu","suffix":""},{"id":399446404,"identity":"64728341-22bb-4e36-b41d-341e62e9425d","order_by":6,"name":"Nan Yang","email":"","orcid":"","institution":"Harin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Nan","middleName":"","lastName":"Yang","suffix":""},{"id":399446405,"identity":"f3ef2ae9-4235-41c4-a77c-ac26bb7eff55","order_by":7,"name":"Jianqin Cao","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAApklEQVRIiWNgGAWjYFCCA2xAwoaHn7+BNC1pMpIzDhBvDUjLYRuDhgQi1fM3Ht726GbbeR4DhgOMHz7mEKFF4sCxcuPctts85swNzJIztxGhxYDhjJk0SItlwwE2Zl4StJzjMTiQQJqWAyRoAfqlTDrnXDKP5IyDzcT5hX/G4W3SOWV29vz8zQc/fCRGC9AaAyiLsYEY9SBrGgwIKxoFo2AUjIKRDQAgdTX/UsTFswAAAABJRU5ErkJggg==","orcid":"","institution":"Harin Medical University","correspondingAuthor":true,"prefix":"","firstName":"Jianqin","middleName":"","lastName":"Cao","suffix":""}],"badges":[],"createdAt":"2025-01-08 02:08:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5784879/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5784879/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":73458226,"identity":"ebe24aa5-0743-4e7b-aef1-0203fe300fa8","added_by":"auto","created_at":"2025-01-10 07:32:24","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":45130,"visible":true,"origin":"","legend":"\u003cp\u003eExperiment paradigm. A: Neutral trial; B: Negative trial.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-5784879/v1/add1826e8046f69430a9d712.png"},{"id":73458212,"identity":"aa9457b0-a4f9-42a2-ab93-55af83ed9b87","added_by":"auto","created_at":"2025-01-10 07:32:23","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":77623,"visible":true,"origin":"","legend":"\u003cp\u003eThe results of behavior data. A: The accuracy in the HG, RG and AG. (B) The response time in the HG, RG and AG. NC= congruent in neutral stimuli; NI= incongruent in neutral stimuli; NCg= congruent in negative stimuli; NIg= incongruent in negative stimuli.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-5784879/v1/3c6766a39527dbdd9c3fd9e2.png"},{"id":73459627,"identity":"27cecf9d-bb5e-4ddd-8201-47f17466d682","added_by":"auto","created_at":"2025-01-10 07:40:23","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":403072,"visible":true,"origin":"","legend":"\u003cp\u003eN2 amplitude evoked by HG, RG and AG. A= neutral stimuli; B= negative stimuli.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-5784879/v1/834a5982367fedbbfc22c597.png"},{"id":73458224,"identity":"9955c977-ab0e-412b-a9c3-1e26ca3fd44e","added_by":"auto","created_at":"2025-01-10 07:32:24","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":402809,"visible":true,"origin":"","legend":"\u003cp\u003eP3 amplitude evoked by HG, RG and AG. A= neutral stimuli; B= negative stimuli.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-5784879/v1/6bc5e2bf9b4cd10c072e5b9b.png"},{"id":73461349,"identity":"cdf94980-3831-42de-9b24-558a0de0bcea","added_by":"auto","created_at":"2025-01-10 07:56:24","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1564848,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5784879/v1/780cb4a9-d6be-468a-94f2-1468f2d449d1.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Deep learning approach to predict developmental outcomes of non-suicidal self-injury: An ERP study","fulltext":[{"header":"Background","content":"\u003cp\u003eNon-suicidal self-injury (NSSI) refers to deliberate and direct damage to body tissue without suicidal intent [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. While not intended to result in death, evidence suggests that NSSI can predict future suicide attempts and fatalities [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Lifetime prevalence rates of NSSI among Chinese adolescents have been reported to range from 16\u0026ndash;29% [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. The classification of NSSI as an independent category in the DSM-5 underscores its significance and the necessity for further research.\u003c/p\u003e \u003cp\u003eNSSI typically begins in early adolescence and progresses into diverse developmental trajectories over time. Group-Based Trajectory Models have identified distinct patterns, such as high-risk and low-risk groups [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Another study involving 13- to 19-year-olds identified three patterns of NSSI frequency: Low Stable, Moderate Fluctuating, and High Persistent [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Similarly, a year-long study of 3,381 Chinese adolescents found two trajectory groups: a faller group (87.4%) and a higher group (12.7%) [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. These longitudinal studies highlight the varied developmental pathways of NSSI, emphasizing the importance of tailored prevention and intervention strategies.\u003c/p\u003e \u003cp\u003eWhile identifying trajectories is crucial, determining predictors of developmental outcomes holds even greater importance. Current forecasting methods largely rely on validated scales. From the perspectives of family system theory and cumulative risk, family conflict risk has been identified as a key predictor of NSSI addiction [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Other studies highlight the roles of interpersonal factors, such as unstable relationships and parental criticism, as predictors of high NSSI fluctuations [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Personal factors, including impulsive behaviors, depression [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e], emotion regulation difficulties, self-compassion deficits [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], psychiatric symptoms, and comorbidities [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e], also contribute significantly. However, reliance on subjective self-reports introduces bias and reduces diagnostic accuracy. Objective measurements based on individual characteristics offer a more reliable alternative, presenting a promising avenue for predicting NSSI outcomes with greater precision.\u003c/p\u003e \u003cp\u003eBehavioral findings from various cognitive tasks have provided insights into cognitive control in subgroups of individuals with NSSI. Frequent NSSI has been associated with poorer performance on the Criticism Gambling Task, suggesting that impulsive decision-making in response to criticism may serve as a critical indicator of NSSI progression [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Additionally, high-severity NSSI groups have been found to exhibit working memory deficits, whereas low-severity groups demonstrate impaired inhibitory control [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Electroencephalogram (EEG) devices allow for the collection of stimulus-evoked EEGs, or event-related potentials (ERPs), through specifically designed experimental tasks. Previous studies have reported that adolescents with NSSI, compared to HC, exhibit increased N2 amplitudes [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e] and decreased P3 amplitudes [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], indicating significant cognitive control dysfunction. Further, Zhao et al. identified that adolescents with NSSI displayed smaller N2 and larger P3 responses during negative emotional face stimuli [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. However, no ERP-based studies currently address developmental outcomes of NSSI, highlighting the potential of investigating neurocognitive factors using ERP methodologies.\u003c/p\u003e \u003cp\u003eERPs offer high temporal resolution, making them valuable for studying cognitive processes with millisecond precision [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. The N2 component, located in the anterior central brain region, is indicative of conflict processing, with its amplitude reflecting the degree of cognitive control [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Cognitive processing involves the activation and coordination of multiple brain regions. For instance, the affective Stroop task studied via fMRI has demonstrated activation in regions such as the amygdala, insula, and prefrontal cortex, alongside cognitive networks involving the prefrontal cortex, superior temporal lobe, and insula [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. This study seeks to examine the cognitive processing features of NSSI outcome groups using an affective Stroop paradigm, which is a reliable tool for assessing cognitive functions [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. Given that negative emotions can impair cognitive performance [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e], the affective Stroop task is a classical paradigm used in recent years [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Washburn et al. noted that individuals frequently engaging in NSSI might have a behavioral addiction [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e], sharing neurobiological roots with addiction[\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. Drug addiction is marked by cue reactivity [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e], shown through behavioral and neurobiological methods like ERP. A prior study indicated that adolescents with NSSI might show altered behavioral and brain responses to NSSI cues [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Therefore, a novel affective Stroop paradigm which contains neutral and NSSI clues images as stimuli was adapted to evaluate the effects of emotional valence on cognitive tasks across NSSI outcome groups. The primary objective was to analyze the dynamic neural mechanisms in these groups through ERP analysis.\u003c/p\u003e \u003cp\u003eDeep learning has demonstrated significant capability in automatically learning and extracting features from raw data, making it highly applicable to EEG signal analysis [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Compared to conventional methods, deep learning techniques offer several advantages, including robust feature extraction and representation, end-to-end training, and superior generalization. Among these, EEGNet, a convolutional neural network (CNN) specifically designed for Brain-Computer Interface (BCI) applications [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], has shown exceptional performance in processing task-related EEG datasets. These include applications such as P3 visual-evoked potentials [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e], movement-related cortical potentials (MRCP) [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], and sensory motor rhythms (SMR) [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. In this study, ERP components such as N2 and P3 were extracted from the affective Stroop dataset for comparative analysis using EEGNet. The secondary objective was to predict developmental outcomes of NSSI by employing these ERP components as input features to the EEGNet model.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eParticipants\u003c/h2\u003e \u003cp\u003eParticipants were selected from an ongoing longitudinal study designed to predict developmental outcomes of NSSI. This study tracked the trajectories of NSSI in 1960 freshmen (mean age\u0026thinsp;=\u0026thinsp;18.13\u0026thinsp;\u0026plusmn;\u0026thinsp;1.65) across three waves, each separated by a six-month interval. NSSI was evaluated based on DSM-5 criteria [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e] for the past six months, which includes eight specific behaviors: cutting/stabbing, hair pulling, head banging, biting, cauterizing, scratching, hitting, and rubbing until bleeding occurred. Latent class analysis revealed three subgroups: healthy group (HG) (n\u0026thinsp;=\u0026thinsp;1627, 83.01%), remission group (RG) (n\u0026thinsp;=\u0026thinsp;276, 14.08%), and aggravation group (AG) (n\u0026thinsp;=\u0026thinsp;57, 2.90%). Ninety freshmen were recruited from the longitudinal study, with 30 participants in each subgroup (HG, RG, AG). Eight participants were excluded: two for accuracy below 75%, three for z-scores above 3.29, and three due to excessive EEG artifacts in over 25% of trials. The final sample included 82 participants (mean age 18.17\u0026thinsp;\u0026plusmn;\u0026thinsp;1.73; 76 females): 27 in HG, 26 in RG, and 29 in AG.\u003c/p\u003e \u003cp\u003eEligibility criteria included normal or corrected vision, no color blindness, no history of alcohol or drug abuse, and no mental illness. Those who met these criteria underwent structured clinical interviews based on DSM-5 criteria conducted by counselors for group assignment confirmation. Participants volunteered for the study, provided informed consent, could withdraw at any time, and were compensated with cash. Ethical approval was obtained from the Ethics Committee of Harbin Medical University Daqing Campus, and the study adhered to the ethical standards outlined in the World Medical Association Declaration of Helsinki.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eTask paradigm\u003c/h3\u003e\n\u003cp\u003e All participants were required to complete the affective Stroop task during EEG recording. The task was created using E-prime 3.0 (see Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). It consisted of two blocks, each containing 120 neutral trials and 120 negative trials. The trials were presented in a pseudo-randomized order within each block. The emotional stimuli included 30 neutral pictures selected from the Chinese Affective Picture System (CAPS) [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e] and 30 neutral pictures sourced from NSSI images of the participants. The two sets of emotional pictures differed significantly in valence [F \u003csub\u003e(1, 58)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;136.51, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001] but were similar in arousal [F \u003csub\u003e(1, 58)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;5.17, P\u0026thinsp;=\u0026thinsp;0.21]. The literal stimuli consisted of single words presented in one of four colors (red, yellow, blue, or green). Two experimental conditions were used: the congruent condition, where each color word was displayed in its corresponding color, and the incongruent condition, where each color word was presented in a color other than the one it referred to (e.g., the word \"red\" displayed in yellow).\u003c/p\u003e \u003cp\u003eEach trial began with a white \u0026ldquo;+\u0026rdquo; displayed for 500\u0026ndash;700 ms in the center of the screen on a grey background, followed by an emotional picture presented for 1000 ms, then a blank screen for 600 ms, followed by a color word displayed for 200 ms, and a final blank screen for 1500 ms. Participants were required to determine whether the word and color were congruent. If they were congruent, participants pressed the \"D\" key with their left index finger; if they were incongruent, participants pressed the \"J\" key with their right index finger. Both response time and accuracy were recorded in this study.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eEEG recording and processing\u003c/h3\u003e\n\u003cp\u003eThe electroencephalogram (EEG) data were recorded using a Neuroscan ESI 64-channel system, equipped with an Ag/AgCl electrode cap and the extended international 10\u0026ndash;20 electrode placement. The system was configured with a bandpass filter of 0.05\u0026ndash;100 Hz and a sampling rate of 500 Hz. A left mastoid electrode was used as the online reference, which was later re-referenced offline to the common average [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. Vertical electrooculogram (EOG) was recorded from electrodes placed above and below the left eye, while horizontal EOG was captured from electrodes positioned at the outer canthi. Electrode impedance was kept below 5 kΩ throughout the task.\u003c/p\u003e \u003cp\u003eExperimental conditions, equipment, and physiological responses can introduce interference that affects the collected signals. To minimize these effects, preprocessing steps were applied to the EEG data. EOG blink artifacts were corrected using linear regression estimation [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]. The EEG signals were band-pass filtered between 0.05 Hz and 30 Hz and segmented into 1000 ms epochs, with a 200 ms pre-stimulus baseline. Epochs with voltage deviations exceeding\u0026thinsp;\u0026plusmn;\u0026thinsp;80 \u0026micro;V were excluded from the ERP analysis. Separate averages for the congruent and incongruent conditions were computed for neutral and negative stimuli, respectively. Only trials with correct responses were included in the calculation of average ERPs.\u003c/p\u003e\n\u003ch3\u003eERP analysis\u003c/h3\u003e\n\u003cp\u003eBased on ERP components linked to cognition as reported in previous research [\u003cspan additionalcitationids=\"CR39\" citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e], as well as the characteristics of the ERP waveforms observed in the current study, specific ERP components were selected for further analysis. The amplitudes of the N2 (260\u0026ndash;320 ms; FZ, FCZ, PZ, CPZ, PZ) and P3 (320\u0026ndash;370 ms; F3, F1, FZ, F2, F4, FC3, FC1, FCZ, FC2, FC4) components were examined. For each ERP component, the mean amplitude within the specified time window was calculated.\u003c/p\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis\u003c/h2\u003e \u003cp\u003eA repeated-measures analysis of variance (ANOVA) was conducted to assess accuracy, reaction time (RT), and ERP component amplitude across different subgroups. The analysis included a three-way repeated-measures design with group (healthy, declining, and worsening) as the between-subject factor, and emotion (neutral, negative) and congruency (congruent, incongruent) as within-subject factors. Statistical analyses were performed using repeated-measures ANOVAs with Greenhouse-Geisser corrections at a significance level of 0.05. Significant interaction effects were followed by simple effect analysis, with pairwise comparisons adjusted using the Bonferroni method and an alpha value of 0.05 divided by the number of comparisons.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eDeep learning algorithm\u003c/h2\u003e \u003cp\u003eIn the model, a 64-channel EEG signal was used as the input, and the group served as the output. The model was trained and tested through a deep convolutional layer followed by a deep separable convolutional layer. This network extracts intrinsic characteristic information from the waveform using convolutional kernels, enabling the prediction of developmental outcomes associated with NSSI. The datasets were split in two ways: randomly into three subsets\u0026mdash;60% for training, 20% for validation, and 20% for testing\u0026mdash;ensuring that no participant's data appeared in more than one subset, and employing 5-fold cross-validation.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eBehavioral performance\u003c/h2\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e presents the accuracy and reaction times (RTs) for the different groups.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe accuracies and average reaction times in the different groups (M\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGroup\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eNeutral stimuli\u003c/p\u003e \u003cp\u003eCongruent Incongruent\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eNegative stimuli\u003c/p\u003e \u003cp\u003eCongruent Incongruent\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eAccuracies\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.94\u0026thinsp;\u0026plusmn;\u0026thinsp;0.06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.84\u0026thinsp;\u0026plusmn;\u0026thinsp;0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e0.94\u0026thinsp;\u0026plusmn;\u0026thinsp;0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e0.81\u0026thinsp;\u0026plusmn;\u0026thinsp;0.04\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.94\u0026thinsp;\u0026plusmn;\u0026thinsp;0.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.84\u0026thinsp;\u0026plusmn;\u0026thinsp;0.09\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e0.94\u0026thinsp;\u0026plusmn;\u0026thinsp;0.11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e0.79\u0026thinsp;\u0026plusmn;\u0026thinsp;0.02\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e0.95\u0026thinsp;\u0026plusmn;\u0026thinsp;0.05\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.84\u0026thinsp;\u0026plusmn;\u0026thinsp;0.06\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e0.94\u0026thinsp;\u0026plusmn;\u0026thinsp;0.07\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e0.80\u0026thinsp;\u0026plusmn;\u0026thinsp;0.06\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eReaction time (ms)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e609\u0026thinsp;\u0026plusmn;\u0026thinsp;91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e646\u0026thinsp;\u0026plusmn;\u0026thinsp;101\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e599\u0026thinsp;\u0026plusmn;\u0026thinsp;113\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e624\u0026thinsp;\u0026plusmn;\u0026thinsp;106\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e623\u0026thinsp;\u0026plusmn;\u0026thinsp;111\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e659\u0026thinsp;\u0026plusmn;\u0026thinsp;109\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e606\u0026thinsp;\u0026plusmn;\u0026thinsp;112\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e610\u0026thinsp;\u0026plusmn;\u0026thinsp;122\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e666\u0026thinsp;\u0026plusmn;\u0026thinsp;91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e712\u0026thinsp;\u0026plusmn;\u0026thinsp;97\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e681\u0026thinsp;\u0026plusmn;\u0026thinsp;147\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e714\u0026thinsp;\u0026plusmn;\u0026thinsp;141\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe ANOVA of accuracy revealed a significant main effect of emotion type (F \u003csub\u003e(1,79)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;21.494, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.214), indicating that accuracy was higher for neutral stimuli compared to negative stimuli (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). The main effect of congruency was also significant (F \u003csub\u003e(1, 79)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;682.573, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.898), showing that accuracy was higher for the congruent condition than for the incongruent condition (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA).\u003c/p\u003e \u003cp\u003eThe ANOVA of RTs showed a significant main effect of group (F \u003csub\u003e(2,79)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;4.145, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05, η\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.095), with slower RTs observed in the AG compared to the HG (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). The main effect of congruency was also significant (F \u003csub\u003e(1, 79)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;65.734, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.454), indicating that RTs were slower for the incongruent condition compared to the congruent condition (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eERP results\u003c/h2\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e presents the average amplitude for each ERP component in the different groups.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe average amplitude for each ERP component in the different groups (M\u0026thinsp;\u0026plusmn;\u0026thinsp;SD)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026plusmn;\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eComponent\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGroup\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eNeutral stimuli\u003c/p\u003e \u003cp\u003eCongruent Incongruent\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eNegative stimuli\u003c/p\u003e \u003cp\u003eCongruent Incongruent\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eN2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e1.09\u0026thinsp;\u0026plusmn;\u0026thinsp;0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e-0.41\u0026thinsp;\u0026plusmn;\u0026thinsp;0.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e-0.41\u0026thinsp;\u0026plusmn;\u0026thinsp;0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e-0.27\u0026thinsp;\u0026plusmn;\u0026thinsp;.81\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e-2.16\u0026thinsp;\u0026plusmn;\u0026thinsp;0.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e-1.66\u0026thinsp;\u0026plusmn;\u0026thinsp;0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e-3.70\u0026thinsp;\u0026plusmn;\u0026thinsp;0.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e-3.63\u0026thinsp;\u0026plusmn;\u0026thinsp;0.83\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e-3.84\u0026thinsp;\u0026plusmn;\u0026thinsp;0.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e-3.98\u0026thinsp;\u0026plusmn;\u0026thinsp;0.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e-3.30\u0026thinsp;\u0026plusmn;\u0026thinsp;0.80\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e-2.89\u0026thinsp;\u0026plusmn;\u0026thinsp;0.78\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eP3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e4.12\u0026thinsp;\u0026plusmn;\u0026thinsp;0.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e2.10\u0026thinsp;\u0026plusmn;\u0026thinsp;0.82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e2.07\u0026thinsp;\u0026plusmn;\u0026thinsp;0.78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e1.39\u0026thinsp;\u0026plusmn;\u0026thinsp;0.76\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e1.05\u0026thinsp;\u0026plusmn;\u0026thinsp;0.77\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e0.19\u0026thinsp;\u0026plusmn;\u0026thinsp;.83\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e-1.98\u0026thinsp;\u0026plusmn;\u0026thinsp;0.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e-2.87\u0026thinsp;\u0026plusmn;\u0026thinsp;0.78\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAG\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c3\"\u003e \u003cp\u003e-0.25\u0026thinsp;\u0026plusmn;\u0026thinsp;0.73\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c4\"\u003e \u003cp\u003e-1.20\u0026thinsp;\u0026plusmn;\u0026thinsp;0.79\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c5\"\u003e \u003cp\u003e-1.43\u0026thinsp;\u0026plusmn;\u0026thinsp;0.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026plusmn;\" colname=\"c6\"\u003e \u003cp\u003e-1.44\u0026thinsp;\u0026plusmn;\u0026thinsp;0.74\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eN2 amplitude\u003c/h2\u003e \u003cp\u003eThe results, presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, revealed a significant main effect of emotion type (F \u003csub\u003e(1, 79)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;7.271, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01, \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\eta\\:\\)\u003c/span\u003e\u003c/span\u003e\u003csup\u003e2\u003c/sup\u003e = 0.084), where the N2 elicited by negative stimuli was larger than that elicited by neutral stimuli. The main effect of group was also significant (F \u003csub\u003e(2, 79)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;7.091, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01, η\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.152). Specifically, the N2 in the HG was significantly smaller than in the RG and AG (ps\u0026thinsp;\u0026lt;\u0026thinsp;0.05). The interaction between group and emotion was significant (F \u003csub\u003e(2, 79)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;16.934, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.300). Under neutral stimuli, the N2 in HG was significantly smaller than in the RG (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05) and AG (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), with the RG also being significantly smaller than the AG (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). Under negative stimuli, the N2 in HG was significantly smaller than in the RG (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) and AG (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05).\u003c/p\u003e \u003cp\u003eThe interaction between group, emotion, and congruency was significant (F \u003csub\u003e(2, 79)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;4.068, p\u0026thinsp;\u0026lt;\u0026thinsp;0.05, η\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.093). Under neutral stimuli, the N2 in HG was smaller than in AG for both congruent (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001) and incongruent (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) conditions, and the N2 in HG was also smaller than in RG for the congruent condition (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). Under negative stimuli, the N2 in HG was smaller than in RG for both congruent (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) and incongruent (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) conditions.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eP3 amplitude\u003c/h2\u003e \u003cp\u003eAs shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, the main effect of emotion type was significant (F \u003csub\u003e(1, 79)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;73.858, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.483), with the P3 elicited by neutral stimuli being larger than that elicited by negative stimuli. The main effect of group was also significant (F \u003csub\u003e(2, 79)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;7.607, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.161), where the P3 in HG was significantly larger than in RG (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) and AG (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01). The main effect of congruency was significant (F \u003csub\u003e(1, 79)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;15.61, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, η\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.165), with the P3 elicited by the congruent condition being significantly larger than that elicited by the incongruent condition (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e \u003cp\u003eNo significant interaction effect between group, emotion, and congruency was observed (F \u003csub\u003e(2, 79)\u003c/sub\u003e\u0026thinsp;=\u0026thinsp;1.411, p\u0026thinsp;\u0026gt;\u0026thinsp;0.05, η\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.034). Simple effects analysis revealed that, under neutral stimuli, the P3 elicited by the congruent condition in HG was larger than in AG (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Under negative stimuli, the P3 elicited by the congruent condition in HG was larger than in RG (p\u0026thinsp;\u0026lt;\u0026thinsp;0.01) and AG (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05), while the P3 elicited by the incongruent condition in HG was larger than in RG (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003ePrediction results on the EEGNet\u003c/h2\u003e \u003cp\u003eWe identified four ERP components during neutral and negative stimulation using the EEGNet model. Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e presents the classification accuracies for N2, P3, and LPP with respect to anxiety under the two affective stimuli. Under neutral stimuli, the ERP components achieved peak accuracies of 94.31% for N2 and 89.60% for P3, while under negative stimuli, the accuracies were 73.16% for N2 and 62.57% for P3. These results indicate that N2 performs best under neutral stimuli, with all components showing reduced accuracy under negative stimuli.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussions","content":"\u003cp\u003eThis study explored cognitive function in college students with varying NSSI developmental outcomes using an affective Stroop task, supporting the role of cognitive function as a predictor of NSSI outcomes. Specifically, the AG exhibited slower reaction times and larger N2 and smaller P3 amplitudes for neutral stimuli compared to RG and HG. The N2 elicited by neutral stimuli, which is associated with identifying developmental outcomes in NSSI, demonstrated high reliability.\u003c/p\u003e \u003cp\u003eOur research utilized neutral and negative images with distinct valence differences as stimuli in the affective Stroop task. The results indicated that negative stimuli impaired cognitive performance relative to neutral stimuli, leading to slower RTs and lower accuracy. This suggests that negative emotions should be closely monitored in individuals with NSSI to prevent adverse outcomes related to impaired cognitive response [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]. The AG exhibited slower RTs compared to the HG, suggesting lower processing efficiency and impaired cognitive control. Participants in the AG required more time to perform conflict detection and task-related activities. No significant differences in accuracy were observed between groups, indicating that NSSI primarily affected processing efficiency rather than task effectiveness, although this finding requires further confirmation due to the small sample size.\u003c/p\u003e \u003cp\u003eNegative stimuli evoked greater N2 amplitudes than neutral stimuli, demonstrating that negative stimuli can rapidly and automatically capture attention and utilize attentional resources. Yuan et al. reported that individuals display heightened sensitivity to negative stimuli, processing them at an earlier stage [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]. The results showed a significant difference in N2 amplitudes among the HG, RG, and AG, with the amplitude exhibiting a gradient change pattern across these groups. These findings indicate that participants in RG and AG, particularly AG, exhibit abnormalities in cognitive control mechanisms. Hanslmayr et al. observed that individuals responded to increased conflict levels by temporarily enhancing cortical responses to task-relevant information, which improved cognitive control performance instead of suppressing task-irrelevant information [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]. Participants in RG and AG were not required to inhibit or overcome interference caused by affective stimuli. This absence of inhibition enhanced attention to target stimuli, resulting in larger N2 components. In contrast, HG displayed superior cognitive control, allowing effective inhibition of task-irrelevant information and successful interference management during task completion [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. Under neutral stimuli, participants in AG exhibited more pronounced N2 amplitude deficits compared to RG, with significant deficits in incongruent N2 amplitude relative to HG. These results suggest that N2 amplitude serves as a sensitive marker for distinguishing HG from NSSI groups, as well as between RG and AG, providing an effective measure for predicting NSSI outcomes.\u003c/p\u003e \u003cp\u003eThe P3 amplitudes were markedly smaller under negative stimuli or incongruent conditions, aligning with findings from previous studies [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]. The P3 component is widely utilized as a biomarker in research on cognitive resource allocation associated with emotion and cognition [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]. It is instrumental in assessing inhibitory processes, such as in the emotional signal stop task, where larger P3 amplitudes signify stronger inhibitory control [\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e]. During conflict processing, suppressing responses to sad faces has been shown to reduce P3 amplitude [\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e]. In the present study, the decreased P3 amplitude indicates that cognitive resources are heavily consumed by negative stimuli or conflicts during the Stroop task. Additionally, the P3 exhibited a gradient downward trend across the HG, RG, and AG, suggesting a progressive decline in the brain's resource utilization efficiency in the NSSI group [\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e]. Under neutral and negative stimuli, AG participants demonstrated more severe P3 amplitude deficits than HG in congruent conditions. Similarly, RG participants displayed significant P3 amplitude deficits under negative stimuli, indicating notable cognitive control impairments compared to HG. These results highlight the importance of continued monitoring for RG participants, even though their NSSI behaviors are alleviated.\u003c/p\u003e \u003cp\u003eNotably, AG participants showed an intermediate pattern under negative stimuli, with N2 and P3 amplitudes positioned between those of HG and RG. This may suggest that negative images become desensitized for AG participants due to NSSI behaviors, resulting in reduced vigilance compared to HG [\u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e]. Alternatively, AG participants may experience more pronounced cognitive dysfunction, requiring additional resources to process negative stimuli, thereby impairing Stroop task performance [\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e, \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e51\u003c/span\u003e]. Although the precise mechanisms remain unclear, these findings provide valuable insights into potential developmental trajectories.\u003c/p\u003e \u003cp\u003e In the RG and AG subgroups, a significant difference was observed in N2 amplitude but not in P3 amplitude, suggesting that AG participants experience greater difficulties with cognitive control over stimuli rather than resource allocation. This finding highlights impaired cognitive control as a central factor in the cognitive dysfunction associated with NSSI development, corroborating earlier studies [\u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e52\u003c/span\u003e, \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e]. The EEGnet model used in this study applies depthwise separable convolutional layers to optimize EEG feature extraction, encompassing spatial filtering and filter bank construction. Notably, the analysis demonstrated that N2 amplitudes elicited by neutral stimuli achieved the highest classification accuracy (94.31%) for predicting developmental outcomes of NSSI among the three ERP components evaluated. These findings establish N2 amplitudes as reliable indicators for identifying developmental outcomes of NSSI, contributing to early detection and intervention strategies. This study suggests that N2 amplitude could serve as a biomarker for predicting NSSI outcomes, offering a potential avenue for developing interventions to reduce addiction risk. Improving cognitive control by addressing deficits in the N2 component may effectively mitigate the risk of addiction in individuals with NSSI.\u003c/p\u003e \u003cp\u003eThis study has several limitations. First, the relatively small sample size may influence the reliability and robustness of the findings. Second, the participants were exclusively college students, which may restrict the generalizability of the results to broader populations. Third, the follow-up period for individuals with NSSI was limited to one year, which may be insufficient for comprehensively evaluating developmental outcomes. A longer follow-up period could potentially reveal different developmental trajectories of NSSI.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eThis study employed a novel affective Stroop task to examine the neurocognitive mechanisms underlying different NSSI outcomes and to predict these outcomes using ERP components. The findings demonstrated that NSSI is associated with abnormal cognitive processing, characterized by impaired cognitive control and resource allocation. Notably, N2 amplitudes emerged as reliable indicators for identifying developmental outcomes of NSSI. These results provide an objective foundation for identifying NSSI, predicting its developmental progression, and formulating early intervention strategies.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eInformed consent was obtained from all participants in the study. This study was approved by the Ethics Committee of Harbin Medical University (HMUDQ202201585).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors thank all the students and schools who participated in this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFei Yin conceived of the study, participated in its design and performed the statistical analysis coordination and drafted the manuscript; Feng Si participated in the design and interpretation of the data; Wenlong Jiang conceived of the study, and participated in its design and helped to draft the manuscript. Shuhui Huo, Binquan Wang, Li Liu and Nan Yang participated in the design and coordination of the study and performed the measurement; Jianqin Cao revised paper and polished language. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research was supported by the National Natural Science Foundation of China (Grant No. 72204065) and the Philosophy and Social Science Foundation of Heilongjiang Provincial (Grant No. 21SHC216).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets produced in the course of this study are not publicly accessible in order to protect the privacy of the subjects. However, they can be obtained from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors report no conflict of interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eNock MK. Self-injury. Annu Rev Clin Psychol. 2010;6:339\u0026ndash;63.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRibeiro JD, et al. Self-injurious thoughts and behaviors as risk factors for future suicide ideation, attempts, and death: a meta-analysis of longitudinal studies. Psychol Med. 2016;46(2):225\u0026ndash;36.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWolff JC, et al. Emotion dysregulation and non-suicidal self-injury: A systematic review and meta-analysis. Eur Psychiatry. 2019;59:25\u0026ndash;36.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiang K, et al. Nonsuicidal self-injury behaviour in a city of China and its association with family environment, media use and psychopathology. Compr Psychiatry. 2022;115:152311.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao H, et al. Cyberbullying victimization and nonsuicidal self-injury in adolescents: Testing a moderated mediating model of emotion reactivity and dispositional mindfulness. J Affect Disord. 2022;299:256\u0026ndash;63.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi L, Yang H. Heterogeneity in Adolescents' Non-Suicidal Self-Injury Behaviour Trajectories Based on the Group-Based Trajectory Model and a Decision Tree Analysis of Family-Related Determinants. Psychol Res Behav Manag. 2023;16:3359\u0026ndash;71.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMason GE, Auerbach RP, Stewart JG. Predicting the trajectory of non-suicidal self-injury among adolescents. J Child Psychol Psychiatry, 2024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang B, et al. Developmental Trajectories of Nonsuicidal Self-Injury in Adolescence and Intrapersonal/Interpersonal Risk Factors. J Res Adolesc. 2017;27(2):392\u0026ndash;406.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFarrell BCT, Ewing L, Hamza CA. Examining trajectories of nonsuicidal self-injury across the first year of university. J Affect Disord. 2024;367:202\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAllen KJD, et al. Frequency of nonsuicidal self-injury is associated with impulsive decision-making during criticism. Psychiatry Res. 2019;271:68\u0026ndash;75.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFikke LT, Melinder A, Landr\u0026oslash; NI. Executive functions are impaired in adolescents engaging in non-suicidal self-injury. Psychol Med. 2011;41(3):601\u0026ndash;10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWen Y, et al. Cognitive Impairment in Adolescent Major Depressive Disorder With Nonsuicidal Self-Injury: Evidence Based on Multi-indicator ERPs. Front Hum Neurosci. 2021;15:637407.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu H, et al. Prefrontal cortex neural activity predicts reduction of non-suicidal self-injury in adolescents with major depressive disorder: An event related potential study. Front Neurosci. 2022;16:972870.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao L, et al. Changes in emotion-related EEG components and brain lateralization response to negative emotions in adolescents with nonsuicidal self-injury: An ERP study. Behav Brain Res. 2023;445:114324.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi F, et al. The Time-Varying Networks in P300: A Task-Evoked EEG Study. IEEE Trans Neural Syst Rehabil Eng. 2016;24(7):725\u0026ndash;33.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZinchenko A, et al. Emotion and goal-directed behavior: ERP evidence on cognitive and emotional conflict. Soc Cogn Affect Neurosci. 2015;10(11):1577\u0026ndash;87.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeidlmayr K, Kihlstedt M, Isel F. A review on the electroencephalography markers of Stroop executive control processes. Brain Cogn. 2020;146:105637.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRaschle NM, et al. Investigating the Neural Correlates of Emotion-Cognition Interaction Using an Affective Stroop Task. Front Psychol. 2017;8:1489.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eScarpina F, Tagini S. The Stroop Color and Word Test. Front Psychol. 2017;8:557.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUher R, et al. Increasing cognitive load reduces interference from masked appetitive and aversive but not neutral stimuli. PLoS ONE. 2014;9(4):e94417.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen L, et al. Impact of acute stress on human brain microstructure: An MR diffusion study of earthquake survivors. Hum Brain Mapp. 2013;34(2):367\u0026ndash;73.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFlaisch T, Steinhauser M, Schupp HT. Adaptive cognitive control attenuates the late positive potential to emotional distractors. NeuroImage. 2019;200:51\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDragan WU, et al. Temperament and neural activation during the affective Stroop task: A functional connectivity study. Pers Indiv Differ. 2022;186:111385.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWashburn JJ, et al. Measuring the urge to self-injure: preliminary data from a clinical sample. Psychiatry Res. 2010;178(3):540\u0026ndash;4.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLuijten M, et al. Systematic review of ERP and fMRI studies investigating inhibitory control and error processing in people with substance dependence and behavioural addictions. J Psychiatry Neurosci. 2014;39(3):149\u0026ndash;69.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDrummond DC. What does cue-reactivity have to offer clinical research? Addiction. 2000;95(Suppl 2):S129\u0026ndash;44.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBurke TA, et al. Emotional response inhibition to self-harm stimuli interacts with momentary negative affect to predict nonsuicidal self-injury urges. Behav Res Ther. 2021;142:103865.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHo CSH, et al. Interpretable deep learning model for major depressive disorder assessment based on functional near-infrared spectroscopy. Asian J Psychiatr. 2024;92:103901.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTian X, et al. Social anxiety prediction based on ERP features: A deep learning approach. J Affect Disord. 2024;367:545\u0026ndash;53.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLawhern VJ, et al. EEGNet: a compact convolutional neural network for EEG-based brain-computer interfaces. J Neural Eng. 2018;15(5):056013.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePolich J. Updating P300: an integrative theory of P3a and P3b. Clin Neurophysiol. 2007;118(10):2128\u0026ndash;48.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGordon SM, et al. Informed decomposition of electroencephalographic data. J Neurosci Methods. 2015;256:41\u0026ndash;55.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLotte F. \u003cem\u003eSignal Processing Approaches to Minimize or Suppress Calibration Time in Oscillatory Activity-Based Brain\u0026ndash;Computer Interfaces.\u003c/em\u003e Proceedings of the IEEE, 2015. 103(6): pp. 871\u0026ndash;890.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAmerican Psychiatric Association, K. Diagnostic and statistical manual of mental disorders: DSM-5. Volume 5. American psychiatric association Washington, DC; 2013.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBai L, Huang MH. The Development of Native Chinese Affective Picture System-A pretest in 46 College Students. Chin Mental Health J. 2005;11(19):719\u0026ndash;22.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKeil A, et al. Committee report: publication guidelines and recommendations for studies using electroencephalography and magnetoencephalography. Psychophysiology. 2014;51(1):1\u0026ndash;21.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSemlitsch HV, et al. A solution for reliable and valid reduction of ocular artifacts, applied to the P300 ERP. Psychophysiology. 1986;23(6):695\u0026ndash;703.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang X, Shangguan C, Lu J. Time course of emotion effects during emotion-label and emotion-laden word processing. Neurosci Lett. 2019;699:1\u0026ndash;7.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFolstein JR, Van Petten C. Influence of cognitive control and mismatch on the N2 component of the ERP: a review. Psychophysiology. 2008;45(1):152\u0026ndash;70.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaetens K, et al. N400 and LPP in spontaneous trait inferences. Brain Res. 2011;1418:83\u0026ndash;92.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHart SJ, et al. Emotional priming effects during Stroop task performance. NeuroImage. 2010;49(3):2662\u0026ndash;70.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYuan J, et al. Are we sensitive to valence differences in emotionally negative stimuli? Electrophysiological evidence from an ERP study. Neuropsychologia. 2007;45(12):2764\u0026ndash;71.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHanslmayr S, et al. The electrophysiological dynamics of interference during the Stroop task. J Cogn Neurosci. 2008;20(2):215\u0026ndash;25.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXinyu G, et al. Weakening of recognition memory caused by task-irrelevant emotional encoding context can be modulated by individuals' inhibitory control. Personality Individual Differences. 2018;134:201\u0026ndash;9.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang K, et al. Extremely negative emotion interferes with cognition: Evidence from ERPs and time-varying brain network. J Neurosci Methods. 2023;396:109922.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVanlessen N, et al. Positive emotion broadens attention focus through decreased position-specific spatial encoding in early visual cortex: evidence from ERPs. Cogn Affect Behav Neurosci. 2013;13(1):60\u0026ndash;79.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSenderecka M. Emotional enhancement of error detection-The role of perceptual processing and inhibition monitoring in failed auditory stop trials. Cogn Affect Behav Neurosci. 2018;18(1):1\u0026ndash;20.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu F, et al. Decreased response inhibition to sad faces during explicit and implicit tasks in females with depression: Evidence from an event-related potential study. Psychiatry Res Neuroimaging. 2017;259:42\u0026ndash;53.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDuncan CC, Kosmidis MH, Mirsky AF. Closed head injury-related information processing deficits: an event-related potential analysis. Int J Psychophysiol. 2005;58(2\u0026ndash;3):133\u0026ndash;57.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaker A, et al. Does habituation matter? Emotional processing theory and exposure therapy for acrophobia. Behav Res Ther. 2010;48(11):1139\u0026ndash;43.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAhumada-M\u0026eacute;ndez F, et al. Affective modulation of cognitive control: A systematic review of EEG studies. Physiol Behav. 2022;249:113743.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu J, et al. Sensitivity to reward and punishment in adolescents with repetitive non-suicidal self-injury: The role of inhibitory control. Int J Clin Health Psychol. 2024;24(2):100456.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu RT. Characterizing the course of non-suicidal self-injury: A cognitive neuroscience perspective. Neurosci Biobehav Rev. 2017;80:159\u0026ndash;65.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Non-suicidal Self-injury, EEG, Developmental outcomes, EEGNet, ERP","lastPublishedDoi":"10.21203/rs.3.rs-5784879/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5784879/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIdentifying predictors of developmental outcomes in non-suicidal self-injury (NSSI) is crucial and goes beyond tracking its progression. EEG technology is notable for its consistent and objective neurophysiological recordings in NSSI detection. Using ERP components in deep learning models for predicting these outcomes is still underexplored.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTwenty-six in the remission group (RG), twenty-nine in the aggravation group (AG), and twenty-seven in the healthy group (HG) completed the affective Stroop task with EEG. N2 and P3 component differences were analyzed across groups, and the EEGNet model was used to assess NSSI developmental outcomes.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResult\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA significant interaction was observed between group and emotion on N2 (F \u003csub\u003e(2, 79)\u003c/sub\u003e = 16.934, p \u0026lt; 0.001, η\u003csup\u003e2\u003c/sup\u003e = 0.300). Under neutral stimuli, N2 was smallest in HG, larger in RG, and largest in AG, while for negative stimuli, N2 in HG was smaller than in RG and AG. A significant group effect on P3 was noted (F \u003csub\u003e(2, 79) \u003c/sub\u003e= 7.607, p \u0026lt; 0.001, η\u003csup\u003e2 \u003c/sup\u003e= 0.161), with HG exhibiting larger P3 compared to RG and AG. The N2 under neutral stimuli achieved the highest classification accuracy (94.31%).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe findings indicate that NSSI is linked to cognitive processing deficits, including impaired control and resource allocation to stimuli. Additionally, N2 amplitudes were shown to reliably predict developmental outcomes in NSSI.\u003c/p\u003e","manuscriptTitle":"Deep learning approach to predict developmental outcomes of non-suicidal self-injury: An ERP study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-01-10 07:32:18","doi":"10.21203/rs.3.rs-5784879/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"c40b2ee2-efff-4373-94c0-af1e4a83932d","owner":[],"postedDate":"January 10th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-01-13T07:23:34+00:00","versionOfRecord":[],"versionCreatedAt":"2025-01-10 07:32:18","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5784879","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5784879","identity":"rs-5784879","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}