Automated EEG Reporting Pipeline for Infant Resting-State Analysis

Manual resting-state EEG reporting for infants is time-consuming and hard for families to interpret. We present an end-to-end pipeline that automates infant EEG reporting and generates family-readable summaries grounded in spectral analysis. The system ingests data, performs preprocessing and artifact handling, and computes power spectral density, band power, and age-normed indices. A lightweight GUI supports full-workflow monitoring and exception handling. To produce plain-language narratives while preserving privacy, we deploy local large language models (Qwen2.5-7B, DeepSeek-V) to condition on structured spectral outputs and render individualized reports. In deployment to 300+ participants from the CBCP cohort, the pipeline delivered age-referenced, frequency-band reports with per-subject textual summaries and reduced report turnaround by ~3 days, improving throughput. This work demonstrates a practical path to scalable, privacy-conserving EEG reporting in early childhood settings and a template for integrating analytical pipelines with on-device language models for clinical communication.