Abstracts

Rationale: The highly sensitive MEG sensors also record environmental noise and contaminants arising from the patient. Real-time methods for noise removal are useful to observe apparent interictal spikes and other abnormal neural activity in the raw data, and equally important, while the patient is present, to detect and possibly remove non-cerebral artifacts that may make detailed post-processing more difficult. The Signal Space Projection (SSP) noise removal built into our MEG system is not readily accessible for direct alteration by us. We needed a process for calculating and updating the SSP basis set in near real-time. Methods: Before the patient arrives, at least two minutes of background activity are recorded from the empty magnetically shielded room. Using mne_browse_raw, a free public research software tool developed by Matti Hämäläinen, we visually examine the data for defects, calculate a new SSP basis set, and save the set in the patient’s record in a format compatible with the vendor’s equipment. The basis set is then copied to the vendor’s specific directory, where it is automatically loaded before each recording. The patient session is broken up into consecutive recording segments, with minimal breaks between segments to minimize the chance of missing an abnormal neural event. In the first segment, the SSP-filtered data are observed in real-time to qualitatively judge if the patient has introduced substantial new artifacts. If so, the recording segment is halted after at least two minutes and the second recording session begun. The first segment is processed as above, and within a few minutes a new SSP basis set is built around the patient’s artifacts and copied to the appropriate directory. The second segment is then halted, and the third and subsequent segments are recorded with the patient-specific basis set. Results: The empty room SSP sets acquired before each patient’s arrival ensured that our environmental noise rejection was appropriately retuned each time to generally unknown noise sources that comprise an urban hospital. This SSP basis set made subsequent identification of artifacts more certain to be arising from the patient. Many patients showed no additional biological contaminants in their data. In several patients, however, magnetic contaminants in their body greatly exaggerated their normal respiratory movements, making observation of their real-time data nearly impossible in the displays. Their adapted SSP basis sets from the first data segment dramatically reduced the biological artifacts in the real-time displays of the third and subsequent data segments. Conclusions: We have adapted into our MEG workflow an efficient procedure for removing in real-time both environmental and biological noise for each patient, by using a research software tool and by changing our MEG recording session practice. Because interictal spikes are rare, the patient-based SSP basis sets generally will not block the interesting abnormal activity in the real-time displays. The SSP method is also non-destructive to the actual recorded data and therefore of no impact to subsequent off-line post processing.