Theodore W. Berger, PhD
David Packard Professor of Engineering
Professor of Biomedical Engineering and Neuroscience
University of Southern California
Berger will present the latest stages in the development of a hippocampal neural prosthesis for declarative memory being tested in humans. This project represents a collaboration between the University of Southern California and Wake Forest University. Epilepsy patients were implanted with depth electrodes in the hippocampus for recordings from CA3 and CA1 single unit spikes during delayed non-match to sample memory testing. “Sample” (1 image) and “match” (3-5 images) stimuli were either CANTAB images or clipart images; delays between “sample” and “match” presentations were either 1-75 sec or 30-75 min for testing short-term memory or long-term memory, respectively. On the initial day of memory testing, during the sample presentations, recordings were obtained from both CA3 and CA1, and from a total of approximately 50-100 neurons from all sites during 100 trials. These data were used to develop a multi-input, multi-output (MIMO) model that allowed CA1 output spike trains to be predicted on the basis of CA3 input spike trains, with input-output relations being specific for each CANTAB or clipart image. On subsequent 1-2 days of testing, some sample trials were “recording only”, some were “MIMO stimulation” (electrical stimulation through the same electrodes otherwise used for recording, with the spatio-temporal pattern being identical to the MIMO model prediction), while others were “random stimulation” (electrical stimulation with the same number of pulses used for MIMO stimulation but with randomized inter-pulse intervals). The overall assumption was that memories are coded in terms of spatio-temporal patterns; the overall prediction was that electrical stimulation during sample presentations with MIMO model-predicted patterns would enhance memory performance during the match phase.
Results for short-term memory (10 patients) showed that electrical stimulation with MIMO-model predicted patterns enhanced memory performance by an average of 23%, with a 47% reduction in error rate; random stimulation produced no change in memory performance. Results for long-term memory (6 patients) showed that electrical stimulation with MIMO-model predicted patterns enhanced memory performance by an average of 30%, with a 44% reduction in error rate; again, random stimulation produced no change in memory performance. Additional studies directed at “decoding” spatio-temporal patterns of pyramidal cell activity have shown promise for being able to read features and categories of hippocampal memory.