(J-376) Predicting Activity States from Phase-Amplitude Coupling in Vim Local Field Potentials

Introduction:: Deep Brain Stimulation (DBS) has proven to be an effective treatment for a variety of medically refractory movement disorders, including Essential Tremor (ET). Optimal tremor control must be balanced with side effects of stimulation, including speech changes which can hinder communication and cause detrimental downstream effects for patients. Current DBS treatment standards deliver continuous stimulation and do not account for patients’ activity states. Differentiating between upper-limb movements and motor activity associated with speech would allow for closed-loop DBS parameter switching to optimize treatment for both conditions. Prior studies have shown the ability to detect movement intent directly from local field potentials (LFPs) of the ventralis intermedius nucleus (Vim). This work explores whether neural signatures of speech and hand movement can be discriminated using LFPs.

Materials and Methods:: LFPs were recorded from bilateral Vim nuclei from three patients with ET during their post-operative DBS mapping appointments at VCU Health, using the Medtronic PERCEPT. The voltage difference between each adjacent contact pair was recorded, providing three signals from four contacts in each hemisphere. The participants were recorded while they were resting, moving both hands, and speaking. Ten seconds from each activity state were split into 100 ms epochs. For each epoch, the modulation index was computed for sixteen pairings: theta (4-8 Hz), alpha (8-13 Hz), low beta (13-21 Hz), and high beta (21-30 Hz) phase bands crossed with four divisions of the gamma band. A logistic regression model with L1 regularization was evaluated using 4-fold cross-validation for each activity state pair.

Results, Conclusions, and Discussions:: Results & Discussion. The average cross-validated testing accuracy across the three participants was 0.63, 0.62, and 0.65 for movement vs. no movement, silence vs. speaking, and movement vs. speaking, respectively. Using a two-sided t-test between the cross-validation accuracies and the accuracy of a model always choosing the dominant class, two of the participants had a testing accuracy significantly above chance for movement vs. still (p = 0.015, 0.007). Tw of the participants had a testing accuracy significantly above chance for silence vs. speaking (p = 0.043, 0.038) and two of the participants had a testing accuracy significantly above chance for movement vs. speaking (p = 0.01, 0.018).

Conclusion. These results suggest speech can be differentiated from periods of silence and hand movement from LFPs, using a simple classifier that would be amenable to real-time deployment. This implies differing representations of motor activity involved in hand motion and speaking in Vim LFPs. The ability to automatically detect different activity states is important for fine tuning the delivery of DBS, as activity-specific stimulation settings could be activated with effective timing. It will be important to further validate model performance with data from additional patients and other neural features.


Acknowledgements (Optional): :

References (Optional): :