Post-processing of a distributed source method for the localization of somatosensory cortex in a cohort of epilepsy patients

Localizing eloquent cortices is crucial for many neurosurgical applications, such as epilepsy and tumor resections. Clinicians may use non-invasive methods such as magnetoencephalography (MEG) to localize these cortical regions using equivalent current dipoles (ECDs). While dipoles are clinically validated, they provide the estimated strength, location, and orientation of only one or a few sources that best describe the recorded neuromagnetic data, requiring clinicians to make subjective decisions on the spatial extent of the underlying cortical area. More accurate delineation of eloquent cortical areas using distributed source localization methods would provide additional pre-surgical information on these regions’ location and spatial distribution, which could lead to reduced post-surgical complications associated with damage to or removal of eloquent cortices. Our objective in this paper was to present a method to post-process the distributed source localization results to yield a directly interpretable, distributed region of activation. As a test case, we selected somatosensory stimulation in a retrospective cohort of focal and multi-focal epilepsy patients. Our algorithm performs source localization using a distributed method (sLORETA), followed by post-processing and blind source separation to identify the area and boundary of the cortical tissue that primarily activates in response to somatosensory stimulation. We calculated the statistical significance of localization by comparing the identified region to an anatomical atlas and random chance. While examining patients who received left (upper left, UL) and right (upper right, UR) sided median nerve stimulation, the cortical areas identified by the algorithm were in anatomically appropriate areas with a median overlap of 97.6% and 94.7%, respectively. We observe that our algorithm localized somatosensory responses better than random chance in 57/58 (98%) patients who performed the UL task (p < 10 × 10⁻¹⁰, binomial test) and 49/50 (98%) patients who performed the UR task (p < 10 × 10⁻¹⁰, binomial test). We compared the localization of our algorithm to current clinical methods and found that our algorithm is not inferior to dipole localization. The algorithm can successfully localize somatosensory responses on the cortical surface in anatomically appropriate regions while providing the spatial extent of cortical activation, reducing subjectivity associated with dipole localization.