0060 A Classification Based Generative Approach to Selective Targeting of Global Slow Oscillations during Sleep

Abstract Introduction Interacting with sleep slow oscillations (0.5,15 Hz) is essential to understanding the mechanisms underlying their role in sleep functions, including homeostasis, synaptic reorganization, and memory consolidation. Current approaches to SO-based non-invasive closed loop alternate current stimulation (cl-tACs) focus on timing, and do not differentiate among oscillations based on their space-time profiles. In this study, we leverage machine learning and global optimization to introduce a generative model of brain stimulation protocols targeting Global SOs. Methods We identify a stimulation protocol that matches a Global SO in two stages: first focusing on the electrode montage, and then on the stimulation waveform. Using the sleep night of 22 volunteers acquired in the lab with high density EEG (175 electrodes), we labeled the SOs as Global/non-Global SOs based on their co-detection on the scalp at a small delay (0.4s). We used the current density (CSD) profile of each SO in cortical and subcortical regions to develop a well-trained classifier (based on weighted k-nearest neighbors). We identify as optimal the electrode montage that resulted in CSDs labeled as Global with highest likelihood by the classifier. We then parameterized a sinusoidal waveform and identified optimum parameters with a genetic algorithm that used feature ranking to establish a weighted distance metric to target Global SO profile built from sleep data. We repeat this approach focusing on Global SO dynamics at a slow time scale and a faster time scale (200ms vs 20ms resolution). Results We identified optimal stimulation profiles matching Global SOs at both time scales. At slow time scale, montage selection relied only on CSD near the SO trough and in the reward network and the brainstem. Linear waveforms over-performed sinusoidal waveforms for fit at low time scales. At fast time scale, CSD at both trough and 500ms before the trough were strongly selective for Global vs. non-Global SO. Conclusion Global SOs are a gating mechanism for large-scale neural communication, a necessary substrate for systems consolidation and long-term memory formation. This data-driven method builds on natural sleep data to shape a stimulation protocol that imitates Global SO dynamics and is generalizable to other SO types. Support (if any)