Spatial Propagation, Synchronization, and Short Term Labor Imminence from Multichannel Electrohysterography: A Methodological Study on the Icelandic 16 Electrode Physionet Database

Electrohysterography (EHG) captures uterine electrical activity noninvasively and can support labor triage. The Icelandic 16‑electrode Electrohysterogram Database (EHGDB) on PhysioNet provides 122 multichannel 4×4 abdominal recordings (200 Hz) from 45 women, including pregnancy clinic visits and recordings during labor. To develop a transparent, reproducible baseline pipeline for distinguishing pregnancy from labor using multichannel EHG. Signals were detrended, band‑pass filtered (0.1–3.0 Hz), and downsampled to 20 Hz. Each recording was segmented into contiguous, non‑overlapping 60‑s windows. For each window we extracted six interpretable features (RMS, variance, approximate entropy, median frequency, magnitude‑squared coherence between channels 1–2, and a conduction‑velocity proxy based on the delay between channels 1–5). A cost‑sensitive Random Forest (200 trees) was evaluated using participant‑grouped 5‑fold cross‑validation; decision thresholds were calibrated per fold to prioritize sensitivity. From 7,153 windows (pregnancy: 6,794; labor: 359), the model achieved AUROC=0.759 and AUPRC=0.319, with recall=60.7% and specificity=79.2% at the calibrated operating point. Fixed‑window, lightweight features provide clinically interpretable performance for pregnancy‑versus‑labor triage on the EHGDB and establish a baseline for future work incorporating contraction‑based segmentation and richer spatial propagation/synchronization measures.