Introduction: Verbal working memory (vWM)—the short-term memory associated with language information—is integral to human cognition and is impaired in many neurosurgical diseases. Much of our knowledge of the neural mechanisms of vWM is derived from functional MRI. This modality is limited in temporal resolution which makes drawing mechanistic conclusions challenging. We therefore sought to use intracranial electroencephalography (iEEG) to investigate the neuromechanistic underpinnings of vWM.
Methods: We recorded iEEG activity from 48 epilepsy patients with stereotactically implanted depth electrodes while they performed a verbal working memory task. The task entailed listening to a list of auditory stimuli and holding the list in memory for several seconds (maintenance) that varied parametrically in load (list length). Activity in the 70-150 Hz band was extracted using the Hilbert Transform and served as a local index of neural computation. Non-parametric permutation tests were used to establish electrodes with significant activity in the maintenance epoch, which were used to decode working memory load using linear discriminant analysis.
Results: After excluding electrodes localized to white matter, 1116 out of 9644 electrodes across patients were found to have significant activity. The majority (524 electrodes) localized to 20 specific locations in both hemispheres, including in the hippocampus, temporal lobe, and prefrontal cortex. Initial results of cross-patient decoding across all patients successfully differentiated between trials with high load (7 or greater items) and low load with a mean accuracy of 0.887, sensitivity of 0.876, specificity of 0.8974, and F1 score of 0.888 across 10 cross-validation repetitions.
Conclusion : Verbal working memory load can be accurately decoded from high-gamma activity in the hippocampus, temporal lobe, and prefrontal cortex. Future work investigating the interactions of these regions has the potential to help protect vWM during surgical intervention, provide new targets for neuromodulation, and improve our understanding of human cognition.
.jpg)
