Shuping Tan

Background: Adolescent depression diagnosis currently relies primarily on subjective self-report questionnaires, with a notable lack of objective neurobiological biomarkers. This study aimed to compare the diagnostic performance of traditional psychological scales with a newly developed Multi-view Adaptive Graph Convolutional Network (MVA-GCN) based on resting-state electroencephalography (EEG), and to exploratorily examine the associations between MVA-GCN-derived brain network features and psychological resilience in adolescents. Methods: Resting-state EEG data were collected from 44 adolescents with major depressive disorder (MDD) and 30 healthy controls (HCs). The MVA-GCN model integrated three parallel connectivity views—phase-locking value (PLV), Pearson correlation coefficient (PCC), and phase lag index (PLI)—via an adaptive fusion mechanism. We compared the classification performance of questionnaire-based machine learning models against the MVA-GCN, and further examined the correlations between model-highlighted network features and clinical symptom measures, including depression, anxiety, loneliness, rumination, and resilience, with age and sex included as covariates. Results: Questionnaire-based machine learning models achieved a mean classification accuracy of 86.43%, whereas the MVA-GCN attained 99.84% accuracy in the present dataset. Occipital and fronto-central regions contributed most to the model’s predictions. Increased gamma-band functional connectivity and reduced alpha-band power were identified as potential electrophysiological correlates of group differences. Exploratory correlational analyses at the nominal level ( p < 0.05, uncorrected) suggested potential group-specific patterns in brain–resilience associations. A post-hoc deviation index analysis showed that, within the MDD group, greater deviation from the healthy reference brain–resilience association pattern was significantly correlated with more severe rumination, self-rated depression, clinician-rated depression severity, loneliness, and self-rated anxiety (all FDR q < 0.05). Conclusion: In this exploratory sample, the MVA-GCN demonstrated promising proof-of-concept discriminative capability for adolescent depression compared with traditional self-report scales. However, the brain–resilience association findings were not statistically significant after correction for multiple comparisons and should be interpreted with caution. These results suggest that MVA-GCN-derived network features warrant further investigation in larger, independent cohorts, but do not yet support their use as a clinically validated diagnostic tool.

Sleep perception impairment (SPI) characterized by subjective-objective discrepancies in sleep, is common among patients with depression. Its neurophysiological mechanisms remain unclear. This study investigated associations between polysomnography (PSG)-derived sleep macro- and micro-architecture features and SPI in depressed patients. We enrolled 63 adults (aged 18-65) with DSM-V major depressive disorder. The participants were divided into two groups: the SPI (n = 26) and non-SPI (n = 37). All underwent overnight PSG and completed clinical assessments. We analyzed sleep macro-architecture and EEG micro-architecture, including sleep temporal entropy (STE), reflecting fragmentation of sleep-stage transitions. Logistic and linear regression models assessed predictors of SPI, adjusting for demographics, clinical, and sleep-related covariates. Despite comparable objective sleep duration, SPI patients significantly underestimated their sleep duration based on their post-PSG subjective sleep time estimates, reported poorer subjective sleep quality, exhibited lower EEG total power (median 9.7 vs. 12.5 kµV²; p = 0.003), decreased interhemispheric EEG symmetry (0.50 vs. 0.51; p = 0.02), and elevated high-frequency relative to slow-wave EEG activity. Higher EEG total power (OR = 0.35 per 1000 µV² increase) and greater EEG symmetry (OR = 0.47 per 0.01 increase) independently predicted reduced odds of SPI in adjusted models. EEG-derived biomarkers (spectral power, symmetry, entropy) may differentiate sleep perception phenotypes in depression, offering potential targets for tailored clinical interventions.

Large language model (LLM) systems can now generate complete research manuscripts, yet their reliability in clinical medicine — where citation accuracy and reporting standards carry direct consequences — has not been systematically assessed. We introduce MedResearchBench, a benchmark of three clinical epidemiology tasks built on NHANES data, and use it to evaluate six AI research systems across six quality dimensions. Evaluation combines programmatic citation verification, rule-based reporting compliance checks, and multi-model LLM judging, providing a more discriminative assessment than conventional single-judge approaches. Citation integrity emerged as the decisive quality dimension. Hallucination rates ranged from 2.9% to 36.8% across systems, and a hard-rule threshold on per-task citation scores capped four of six systems’ total scores at the penalty ceiling. Adding a multi-agent citation verification and repair pipeline to the best-performing system improved its citation integrity score from 40.0 to 90.9 and raised the weighted total from 68.9 to 81.8. Strikingly, a single-model evaluation ranked this system last (55.5), while our three-tier framework ranked it first (81.8) —a complete reversal that exposes the limitations of subjective LLM-only evaluation. These results suggest that programmatic citation verification should be a core metric in future evaluations of AI scientific writing systems, and that multi-agent quality assurance can bridge the gap between fluent text generation and trustworthy scholarship.

The rapid advancement of AI research automation systems—including AI Scientist, data-to-paper, and Agent Laboratory—has demonstrated the potential for autonomous scientific discovery. However, existing benchmarks for evaluating these systems focus pre-dominantly on fundamental sciences (machine learning, physics, chemistry), overlooking the unique challenges of medical clinical research: complex survey designs, inferential statistics with confounding control, adherence to reporting standards (STROBE, CONSORT), and the requirement for clinically actionable interpretation. We present MedResearchBench , the first benchmark specifically designed to evaluate AI systems on medical clinical research tasks. MedResearchBench comprises 16 tasks spanning 7 clinical domains (cardiovascular, oncology, mental health, metabolic, respiratory, neurology, infectious disease), built on publicly available datasets (the National Health and Nutrition Examination Survey [NHANES] and the Surveillance, Epidemiology, and End Results [SEER] program) with ground truth from 16 high-quality published papers (IF range: 2.3–51.0). Each task is evaluated along 6 medical-specific dimensions: statistical methodology, results accuracy, visualization quality, clinical interpretation, confounding sensitivity, and reporting compliance. We describe the benchmark design rationale, task construction methodology, paper selection criteria with anti-paper-mill filtering, and a detailed analysis of task characteristics including methodological diversity, evaluation dimension coverage, and difficulty stratification. To demonstrate benchmark executability, we evaluate an agentic data2paper pipeline on 3 pilot tasks spanning all three difficulty tiers, achieving scores of 72/100 (Tier 1, Cardio 000), 69/100 (Tier 2, Mental 000), and 75/100 (Tier 3, Metabolic 002), with a mean score of 72/100 (B-level). Survey-weighted methodology was correctly implemented across all tasks; primary limitations included covariate incompleteness and reference group misspecification. MedResearchBench addresses a critical gap in AI research evaluation and provides a standardized, community-extensible platform for assessing whether AI systems can conduct clinically sound, publication-quality medical research. All task materials are publicly available at https://github.com/TerryFYL/MedResearchBench .

Objective. Motor imagery EEG (MI-EEG) decoding remains challenging due to low signal-to-noise ratios and pronounced inter-subject variability. Although end-to-end deep models reduce reliance on manual feature engineering, many existing architectures may introduce temporal leakage through non-causal operations and often rely on fixed spatial topologies that cannot accommodate subject- and trial-specific connectivity patterns. Approach. We propose MAGCANet, which integrates five core components: (i) a Multiscale Causal Convolution Module (MCCM) for hierarchical temporal encoding under explicit causal constraints, (ii) a Temporal Convolution Module (TCM) to capture complex temporal dynamics, (iii) an Adaptive Graph Convolution Module (AGCM) for sample-specific topology learning in latent space, (iv) a Multi-Head Self-Attention Module (MHSAM) for global feature aggregation, and (v) a Classification Block for final decision making. Together, these components enforce temporal causality, adapt spatial interactions to individual dynamics, and produce discriminative representations robust to inter-subject variability. Results. On the BCI Competition IV-2a and IV-2b datasets, MAGCANet achieves strong single-subject accuracies of 88.58\% and 91.13\%, respectively. Under Leave-One-Subject-Out (LOSO) evaluation, the model maintains accuracies of 70.49\% and 79.49\%, demonstrating competitive and stable cross-subject generalization. MAGCANet is highly lightweight, with only 0.0194M parameters, and achieves low inference latency (2.23 ms). Qualitative analyses, including feature clustering and channel occlusion, further highlight the model’s interpretability and its ability to capture relevant EEG patterns. Significance. MAGCANet provides a robust and interpretable solution for MI-EEG decoding, balancing high precision with computational efficiency, and offering a reliable method for real-time BCI applications.