Optically-pumped magnetometer magnetoencephalography (OPM-MEG) is uniquely sensitive to movement and ambient-field interference. Homogeneous Field Correction (HFC) suppresses the spatially-uniform component of this interference but, being an orthogonal projection, also removes neural signal that overlaps the interference subspace. The recent opHFC method replaces the orthogonal step with an oblique projection to preserve overlapping signal, but requires a per-subject MRI lead-field for its signal subspace and an empty-room recording for its noise subspace. We present AO-HFC (Adaptive Oblique HFC), an oblique-projection HFC that needs neither MRI nor empty-room: the signal subspace is built from a geometric single-sphere (Sarvas) forward model using only sensor positions and orientations, and the interference subspace is estimated data-driven from the recording's own low-frequency structure. An optional time-varying formulation re-estimates the operator in overlapping windows while remaining structurally power-reducing, and a do-no-harm safeguard falls back to standard HFC whenever the oblique step would add band power. On two independent, public OPM datasets spanning two vendors and a wide range of head motion, AO-HFC removes 96–98% of low-frequency interference (vs 54–82% for standard HFC) while preserving brain-band signal comparably under low motion and markedly better under high motion, and never degraded a recording across all windows tested. The software is distributed separately as a sealed container; this report is released to document and timestamp the method. Research use only; not a medical device.
Xuan Dong Nguyen (Sat,) studied this question.