NeuroCon-AutismNet: a privacy-preserving multimodal framework toward autism screening via diffusion-regularized EEG
J Revathy1, Karthiga M2, Sumendra Yogarayan3
1Department of Artificial Intelligence and Data Science, Christ the King Engineering College, Coimbatore, Tamil Nadu, India.
Frontiers in Psychiatry
|August 8, 2026
Summary
This study introduces NeuroCon-AutismNet, a novel multimodal architecture for Autism Spectrum Disorder (ASD) screening, integrating EEG synthesis and multilingual conversational analysis with differential privacy. Current validation uses synthetic data, showing no diagnostic capability yet.
Area of Science:
- Neuroscience
- Machine Learning
- Computational Psychiatry
Background:
- Autism Spectrum Disorder (ASD) screening needs multimodal biomarkers.
- Current methods for ASD screening are siloed (EEG, conversation).
- Privacy-preserving, multilingual frameworks for mental health screening are lacking.
Purpose of the Study:
- To present NeuroCon-AutismNet, a proof-of-concept multimodal architecture for ASD screening.
- To integrate EEG synthesis, multilingual conversational screening, and differential privacy.
- To establish architectural readiness for future real-data deployment.
Main Methods:
- Developed NeuroCon-AutismNet with four modules: TDBG (EEG synthesis), MADSN (multilingual conversation), NLFT (fusion), and AMEL-X (fusion).
- Integrated formal (ε, δ)-differential privacy using DP-SGD.
- Validated architecture using synthetic data and partial external benchmarks.
Main Results:
- Synthetic evaluation showed diagnostic AUC of 0.503, not statistically different from chance.
- EEG spectral comparison yielded Pearson r = 0.87, with poor reproduction in theta and gamma bands.
- Expert evaluation of conversational module (MADSN) showed 90% empathy satisfaction and Cohen's κ = 0.82.
Conclusions:
- The current synthetic evaluation prevents any screening or clinical-utility claims.
- NeuroCon-AutismNet is architecturally ready for future real-data deployment.
- Real-data validation and further studies are prerequisites for clinical utility.
