Social Pain Disrupts Emotion-Action Brain-State Dynamics in Adolescents with Non-Suicidal Self-Injury

Ying Xu; Xiaojun Liang; Li Zhang; Yixuan Yuan; Gan Huang; Yongjie Zhou; Zhen Liang

cs.AIOct 8, 2026

Social Pain Disrupts Emotion-Action Brain-State Dynamics in Adolescents with Non-Suicidal Self-Injury

Authors: Ying Xu, Xiaojun Liang, Li Zhang, Yixuan Yuan, Gan Huang, Yongjie Zhou, Zhen Liang

Organizations: School of Biomedical Engineering, Medical School, Shenzhen University, Shenzhen, China. · Guangdong Provincial Key Laboratory of Biomedical Measurements and Ultrasound Imaging, Shenzhen, China. · Pengcheng Laboratory, Shenzhen, China.

Abstract

Non-suicidal self-injury (NSSI) is prevalent among adolescents with depression, but the rapid brain-state dynamics linking social distress to maladaptive behavior remain unclear. We combine an experimental pain paradigm, electroencephalography (EEG) microstate analysis, and interpretable deep sequence modeling to investigate NSSI-related neurodynamics in 106 adolescents with depression, including 67 with NSSI (DN+) and 39 without NSSI (DN-), during social pain, physical pain, and resting-state conditions. A model integrating disease-specific, domain-adversarial, consistency, and contrastive learning captures higher-order dependencies in microstate sequences. Social pain yields the strongest NSSI discrimination, with 68.55% accuracy, outperforming the best baseline by 8.94% points. Model interpretation and conventional microstate analyses reveal weakened bidirectional transitions between MS3 and MS5 in DN+ adolescents during social pain. Source reconstruction associates MS3 with emotional/interoceptive processing and MS5 with action preparation, suggesting disrupted emotion-action coupling. Time-resolved analyses show greater early-to-middle action-state recruitment and later emotion-state recruitment in DN+ adolescents. In DN- adolescents, MS5-to-MS3 dynamics mediate associations between social-evaluation sensitivity and affective outcomes, whereas this mediation is absent in DN+; conversely, MS3-to-MS5 transitions are associated with greater negative affect in DN+. Together, these findings identify disrupted emotion-action coupling as a key neurodynamic mechanism underlying altered social pain processing in adolescents with NSSI, providing a mechanistically interpretable neural signature for objective identification of NSSI.

Figures & tables

Figure 1: Overview of the proposed model framework. The model consists of four main components.(1) Microstate sequence construction. Preprocessed EEG signals are subjected to a two-step microstate clustering procedure using the k-means algorithm to obtain optimal microstate templates. These templates are then back-fitted to each data segment to generate corresponding microstate time series. (2) MS temporal reorganization. The microstate sequences are segmented into patches using sliding windows (window length = 50, step size = 50). Two shuffling strategies (weak permutation and strong permutation) are further applied to enhance the diversity of temporal patterns and improve model robustness. (3) Neural representation learning. A PatchTST architecture is employed to capture dynamic transition patterns within the microstate sequences and map them into high-dimensional representations. A dedicated CLS patch is introduced to facilitate the integration of local segment-level characteristics and global sequence-level structure. (4) Dynamic stability learning. The model is jointly optimized using multiple objectives, including a disease-specific loss, domain adversarial loss, consistency loss, and contrastive learning loss. This joint optimization enables the model to anchor disease-relevant pathological microstate transition patterns associated with NSSI, thereby samplifying subtle differences between patient subgroups.

State	Accuracy(%)	Precision(%)	Recall(%)	F1-score(%)	AUC(%)
Resting	61.84 $\pm$ 01.39	62.26 $\pm$ 03.10	62.11 $\pm$ 10.00	61.60 $\pm$ 04.13	60.50 $\pm$ 02.64
Physical Pain	66.45 $\pm$ 04.31	67.07 $\pm$ 05.49	66.58 $\pm$ 11.31	66.12 $\pm$ 05.98	66.00 $\pm$ 05.73
Social Pain	68.55 $\pm$ 03.07	69.49 $\pm$ 04.60	67.37 $\pm$ 06.70	68.08 $\pm$ 03.41	68.06 $\pm$ 04.57

Table 1: Model performance across social pain, physical pain, and resting-state conditions

Figure 2: Cosine similarity between model-derived EEG representations and stimulus representations across pain conditions. (a) Group-level comparison of cosine similarity between model-derived EEG features and stimulus image features under social pain (SP) and physical pain (PP) conditions. (b-c) Within-group comparisons of SP versus PP for DN

+

and DN

-

adolescents, respectively. (d) Between-group comparison of cosine similarity across conditions. (e-f) Post hoc comparisons between DN

+

and DN

-

groups under SP and PP conditions, respectively. *p

<

0.05.

Figure 3: Microstate transition attention patterns across clinical and sex subgroups. (a-b) Mean microstate transition attention matrices for DN $+$ and DN $-$ adolescents, averaged across ten-fold cross-validation. (c) Statistical comparison of transition attention weights between clinical groups. (d-e) Mean transition attention matrices for male and female subgroups. (f) Statistical comparison between sex subgroups. Asterisks indicate transitions showing significant group differences. *p $<$ 0.05.

Figure 4: Source-level reconstruction and functional annotation of EEG microstates. Source-level activation patterns of the eight microstate templates are shown together with their meta-analytic functional annotations derived from Neurosynth. For each template, word clouds display the top 15 positively and negatively associated functional terms based on correlation coefficients, excluding anatomical terms. The cortical region showing the strongest source activity for each template is indicated below the corresponding map.

Figure 5: Time-resolved validation of microstate dynamics during social pain processing. A sliding-window approach (window length = 300 ms; step size = 50 ms) is used to quantify the temporal occurrence of each microstate. Microstate templates with identical spatial topographies but opposite polarity were treated as the same state. Solid lines represent adolescents with depression and NSSI (DN $+$ ), whereas dashed lines represent adolescents with depression without NSSI (DN $-$ ). Shaded regions indicate time intervals showing significant between-group differences. Corresponding source-level activation patterns are displayed for each microstate.

Figure 6: Associations among social evaluation, microstate dynamics, and affective outcomes. (a-b) Statistical mediation analyses examining MS5 $\rightarrow$ MS3 dynamics in the association between sensitivity to social evaluation and depressive symptoms (PHQ-9 scores) in DN $-$ and DN $+$ adolescents. (c-d) Corresponding analyses for subjective negative emotional ratings following pain-related stimuli. (e) Association between MS3 $\rightarrow$ MS5 transition probability and subjective negative emotional ratings across groups. Solid red lines indicate statistically significant paths, whereas dashed black lines indicate non-significant paths. Path coefficients are shown alongside each path. DN $+$ , depression with NSSI; DN $-$ , depression without NSSI.

Method	Accuracy(%)	Precision(%)	Recall(%)	F1-score(%)	AUC(%)
KNN [ 12 ]	53.55 $\pm$ 04.47	53.15 $\pm$ 03.87	61.58 $\pm$ 07.87	56.86 $\pm$ 04.77	47.47 $\pm$ 05.78
Decision tree [ 5 ]	47.50 $\pm$ 04.62	47.30 $\pm$ 04.60	49.21 $\pm$ 11.84	47.84 $\pm$ 07.62	48.33 $\pm$ 04.00
SVM [ 11 ]	46.45 $\pm$ 04.96	46.34 $\pm$ 05.25	46.84 $\pm$ 08.02	46.48 $\pm$ 06.23	55.37 $\pm$ 06.17
Random forest [ 6 ]	52.24 $\pm$ 06.30	52.23 $\pm$ 06.07	52.89 $\pm$ 08.08	52.44 $\pm$ 06.63	52.47 $\pm$ 06.91
RNN [ 62 ]	58.03 $\pm$ 05.80	64.11 $\pm$ 10.92	49.74 $\pm$ 26.82	50.72 $\pm$ 15.13	56.11 $\pm$ 09.70
LSTM [ 26 ]	56.45 $\pm$ 04.09	65.08 $\pm$ 10.70	38.95 $\pm$ 24.46	44.27 $\pm$ 11.79	53.36 $\pm$ 05.40

Table 2: Performance comparison with conventional machine-learning, neural sequence, domain-adaptation, and contrastive-learning methods.

Train set	Test set	Accuracy(%)	Precision(%)	Recall(%)	F1-score(%)
Entire data	Entire data	68.55 $\pm$ 03.07	69.49 $\pm$ 04.60	67.37 $\pm$ 06.70	68.08 $\pm$ 03.41
Female subgroup	Female subgroup	67.63 $\pm$ 06.62	67.76 $\pm$ 04.90	67.89 $\pm$ 17.59	66.87 $\pm$ 09.82
Female subgroup	Male subgroup	65.79 $\pm$ 02.79	66.48 $\pm$ 04.36	65.26 $\pm$ 10.43	65.30 $\pm$ 04.74
Male subgroup	Male subgroup	65.79 $\pm$ 02.46	67.35 $\pm$ 06.00	65.79 $\pm$ 16.95	65.00 $\pm$ 06.41
Male subgroup	Female subgroup	66.32 $\pm$ 02.00	72.51 $\pm$ 05.64	53.68 $\pm$ 05.46	61.37 $\pm$ 02.23

Table 3: Sex-specific and cross-sex validation of the proposed framework.

Data	Model	Accuracy(%)	Precision(%)	Recall(%)	F1-score(%)	AUC(%)
HUSM (EO)	KNN	53.80 $\pm$ 10.48	62.90 $\pm$ 28.52	27.17 $\pm$ 14.28	36.05 $\pm$ 15.48	56.54 $\pm$ 18.95
	Decision tree	53.48 $\pm$ 11.06	55.20 $\pm$ 13.84	54.57 $\pm$ 12.49	54.03 $\pm$ 09.47	53.94 $\pm$ 11.48
	SVM	49.13 $\pm$ 04.31	49.02 $\pm$ 05.73	40.43 $\pm$ 07.19	44.07 $\pm$ 06.03	48.31 $\pm$ 04.43
	Random forest	61.96 $\pm$ 19.70	61.12 $\pm$ 21.72	76.96 $\pm$ 23.44	66.75 $\pm$ 19.43	60.07 $\pm$ 24.39
	RNN	63.37 $\pm$ 06.91	58.58 $\pm$ 22.33	64.13 $\pm$ 28.22	59.69 $\pm$ 21.86	60.78 $\pm$ 07.92
	LSTM	62.61 $\pm$ 06.57	62.02 $\pm$ 08.99	74.57 $\pm$ 12.76	66.50 $\pm$ 04.52	60.56 $\pm$ 11.43

Table 4: External validation on independent depression datasets.

Figure 7: Microstate transition attention weight matrices for MDD and HC in public datasets. (a–c) HUSM under the EO conditions, (d–f) HUSM under the EC conditions, and (g–i) MODMA under the EO conditions, showing averaged ten-fold attention matrices and paired t-test results. *p $<$ 0.05.

Figure 8: Experimental paradigm. The experiment consisted of a 5-minute resting-state session and a task session. The task session included two blocks (social pain and physical pain) presented in a randomized order. Each block comprised 35 trials, and each trial consisted of four phases: (1) Preparation phase (2 s), during which participants fixated on a central cross; (2) Stimulus phase (5 s), in which a task-relevant image is presented and participants passively viewed it; (3) Emotion rating phase (up to 5 s), during which participants rated their subjective negative emotional experience using a keyboard (1 = no negative emotion, 9 = most intense negative emotion), with the phase terminating immediately upon response; and (4) Inter-trial interval (1 s). The order of image presentation within each block was randomized across participants.

Explore similar work

Sep 29, 2026cs.NE

NeuroDyn-EEG: An Interpretable Pre-trained Model for EEG Based on Neural Dynamics

Clinical scalp electroencephalography (EEG) offers a noninvasive window into neural dynamics of neuropsychiatric disorders. However, discriminative deep models often lack anatomically indexed physiological interpretability. We propose NeuroDyn-EEG, a pretraining framework integrating generative priors from neural dynamics. It couples an extended Jansen-Rit neural mass model, leadfield-based source projection, and simulation-based parameter inversion. Trained on synthetic parameter-EEG pairs within physiological ranges, NeuroDyn-EEG estimates 11 regional parameter families across 90 AAL regions plus one global parameter from standard 19-channel EEG, using only ~2.43M trainable parameters. We evaluate the framework across three levels. First, controlled simulations demonstrate robust parameter recovery under diverse noise conditions, while real resting-state EEG evaluations confirm spectral and phase consistency in an inverse-forward closed loop. Second, on four clinical benchmarks (AD65, PD31, Figshare MDD, and TUAB), NeuroDyn-EEG achieves competitive classification performance, securing the highest BACC, AUROC, and AUCPR on PD31 and MDD, and highest BACC on AD65. Third, post hoc regional analyses reveal disease-specific alterations: local synaptic connectivity C_1 involves the most altered regions in AD65, whereas the firing threshold theta ranks first in MDD, offering testable mechanistic hypotheses. Overall, NeuroDyn-EEG maps scalp EEG to anatomically indexed dynamical parameters, bridging representation learning and mechanistic neurophysiology. Code: https://github.com/Gnosis-Neurodynamics/NeuroDyn-EEG.

Yi Cui, Tong Zhao, Jiaxin Lei +5

Gnosis Neurodynamics Co. Ltd, Room 202, West Area, Building North 1, No. 9 Yard, Shengmingyuan Road, Changping District, Beijing, China · School of Biomedical Engineering, Tsinghua Medicine, Tsinghua University, Beijing, China · University of California, Davis, department of psychology +2

Electroencephalography Neural Posterior Estimation

Ask PDF

May 17, 2026q-bio.NC

Von Economo neurons enable reliable social skill acquisition in recurrent spiking neural networks: a computational account with clinical predictions

Von Economo neurons (VENs) are selectively lost in behavioural-variant frontotemporal dementia (bvFTD) and reduced in autism spectrum conditions (ASC), yet their computational role in social learning remains unexplained. We train a spiking neural network (the VENCircuit) embedding VEN-like projection neurons (K=40, 2% of total) in a recurrent pyramidal circuit across 50 matched random initialisations with and without VENs. The network is trained on a controlled binary classification task; we make no claim to model social cognition directly. VEN-intact networks converged in 49/50 cases (98%) versus 35/50 (70%) for VEN-ablated networks (Fisher's exact OR=21.0, 95% CI 2.7-167, p=8.7e-5). Failed ablated networks showed complete absence of learning, inconsistent with a speed-of-learning account. Phase-ablation experiments show VEN removal is most disruptive during mid-training (epochs 5-25), when a co-adaptive dependency forms in the pyramidal circuit. We derive a formal account showing VENs provide a direct gradient pathway immune to Jacobian instabilities affecting the recurrent circuit. Inference-time VEN ablation caused a significant performance drop (Wilcoxon p=0.022), ranging from no change (16/20 networks) to catastrophic collapse (0.989 to 0.620). VENs function as acquisition scaffolds whose developmental absence produces stochastic learning failure - a computational analogue of variable social skill acquisition in ASC - with falsifiable predictions for organoid and electrophysiology studies.

Esila Keskin

School of Computing and Creative Technologies, University of the West of England, Bristol, UK

Neural Processes Recurrent Neural Networks Spiking Neural Networks Neural Network Training Dynamics

Ask PDF

Sep 27, 2026cs.LG

T-SNN: Temporal Simplicial Neural Network for EEG Decoding

Decoding brain states requires models that capture both the evolution of neural activity and interactions among groups of brain regions. Existing EEG methods often treat recordings as multivariate time series or represent functional connectivity with pairwise graphs, leaving dynamic higher-order interactions largely unmodeled. We introduce the Temporal Simplicial Neural Network (T-SNN), which represents EEG recordings as sequences of evolving simplicial complexes. By combining simplicial convolutions with recurrent updates, T-SNN jointly learns higher-order interactions and their temporal evolution. On the seven-class SEED-VII emotion recognition task, T-SNN outperforms convolutional, recurrent, graph-based, and Transformer methods in both trial-wise and cross-subject evaluations. Incorporating eye-movement features further improves performance, demonstrating the framework's potential for multimodal brain-state decoding.

Nikita Malik, Shubhajit Roy, Mohit Kataria +4

Bharti School of Telecommunication Technology and Management Indian Institute of Technology, Delhi · Department of Computer Science & Engineering Indian Institute of Technology, Gandhinagar · Yardi School of Artificial Intelligence Indian Institute of Technology, Delhi +4

Cross-Subject EEG Decoding EEG Decoding Multimodal Emotion Recognition EEG Emotion Recognition

Ask PDF