Childhood Mental Health Assessment Model Integrating Intestinal Microbiota Feature Map and Daily Behaviour Identification
Shuo Gu, Fenghua WuAbstract
Objective
Aiming at the clinical limitations of conventional child mental health assessments that rely on subjective scales and suffer from insufficient early identification accuracy, this study constructs a multi-dimensional integrated AI evaluation model on the basis of microbiota-intestine-brain axis theory, providing a new paradigm for neurodevelopmental risk screening in children.
Methods
A total of 235 children aged 3-12 were enrolled in this study when faecal metagenomic sequencing data were collected to establish an intestinal microbiota feature map, from which core features including neuroactive metabolite abundance, microbial diversity indices and relative abundance of differential genera were extracted simultaneously. Their daily behavioural time-series data covering such indicators as social interaction, emotional fluctuation, attention maintenance and stereotyped movements were acquired via an edge-aware system. A multimodal attention mechanism was applied to fuse these two heterogeneous feature sets, where reinforcement learning algorithm was adopted for DeepSeek model training. Internal testification was performed by using k-fold grouped cross-validation, while generalization performance was confirmed through a multi-centre external cohort.
Results
The proposed integrated model achieved an AUC of 0.912 and an accuracy of 89.7% in identifying abnormal neurodevelopment in children, thus demonstrated effective discriminative performance in subgroups of autism spectrum disorder and attention-deficit/hyperactivity disorder.
Conclusion
The integration of intestinal microbiome features and daily behavioural characteristics effectively overcomes the subjective bias limitations of traditional assessments. With the advantages of non-invasiveness and high timeliness, the integrated AI model can provide considerable technical support for early screening and proper intervention of children mental health defects.
Acknowledgements
This research was financially supported by Liaoning Province Applied Basic Research Program Project: Research on Multimodal Embodied Intelligent Perception and Sports Evaluation Technology for Rowing Sculling Pool (1763689282436).
Corresponding Author
Fenghua Wu, Shenyang City University, Shenyang, China.