MEG source models, coregistration and interpretation
MEG source analysis estimates a source explanation for measured sensor fields. The forward problem predicts the measurements from assumed currents and a head model. The inverse problem works in the other direction and is not uniquely solved by the sensor data alone. Assumptions about location, orientation, extent, number, timing and regularization influence what can be reconstructed. A source map is therefore a model-based estimate rather than a direct photograph of neural tissue.
Coregistration connects functional measurements and structural anatomy. Fiducial labels, surface points, head position, MRI identity and the chosen transformation all require review. A small average surface error can conceal a mislabeled landmark or a locally poor fit. Inspect the actual correspondence rather than accepting a single summary number. An excellent sensor-space dipole fit does not repair an incorrect anatomical transformation.
Equivalent-current-dipole fitting can be useful for appropriate focal patterns. Goodness of fit summarizes agreement with the measured field under the chosen model, not proof that the estimated source is epileptogenic or anatomically exact. Distributed approaches and spatial filters have other assumptions and limitations. Correlated sources, covariance estimation, noise, excluded channels and head modeling can affect results. Review sensitivity to reasonable analysis choices rather than select the most attractive location without disclosure.
Preprocessing must preserve traceability. Filtering can alter transient shape, while component removal or trial selection can remove physiological activity as well as artifact. Keep raw data, describe settings and exclusions and ensure that qualified reviewers can reproduce the process. The familiar square-root averaging benefit is conditional; coherent artifact can persist and timing variability can blur an evoked response.
Clinical meaning requires integration with history and other investigations. Concordant source estimates can contribute to presurgical assessment, but a tight cluster does not independently define resection or guarantee seizure freedom. Functional dominance estimates also require task validity, appropriate interpretation and consideration of remaining risk. Display thresholds and colors are not exact disease boundaries. This bank teaches analytical distinctions and cannot replace supervised source analysis, expert clinical reporting or the multidisciplinary decision process.