建模用于脑机接口的电气神经刺激

© 2024 by Ramandeep Singh Vilkhu. All Rights Reserved.Re-distributed by Stanford University under license with the author. This dissertation is online at: https://purl.stanford.edu/yw594wb3690 I certify that I have read this dissertation and that, in my opinion, it is fully adequate inscope and quality as a dissertation for the degree of Doctor of Philosophy. Subhasish Mitra, Primary Adviser I certify that I have read this dissertation and that, in my opinion, it is fully adequate inscope and quality as a dissertation for the degree of Doctor of Philosophy. E.J. Chichilnisky I certify that I have read this dissertation and that, in my opinion, it is fully adequate inscope and quality as a dissertation for the degree of Doctor of Philosophy. Tsachy Weissman Approved for the Stanford University Committee on Graduate Studies. Stacey F. Bent, Vice Provost for Graduate Education Abstract Brain-computer interfaces (BCIs) have the potential to revolutionize medicine by replacing degen-erated neural pathways with embedded electronics.A prominent example is the use of epiretinalimplants to restore vision in people blinded by age-related macular degeneration (AMD) orretinitispigmentosa(RP). In these conditions, the photoreceptors in the retina degenerate, while the reti-nal ganglion cells (RGCs) remain largely intact, providing an opportunity for the reintroduction ofvisual information through electrical stimulation. Natural vision results from precise coordination of electrical signals in RGC sub-types, each ofwhich conveys unique visual features to the brain.Thus, electrically eliciting precise patterns ofRGC activation is likely necessary to restore natural vision. Current devices pass current througha single electrode at a time (single-electrode stimulation), resulting in insufficient activation of tar-geted RGCs or indiscriminate activation of non-targeted RGCs. Stimulation through multiple elec-trodes (multi-electrode stimulation) potentially enables more targeted stimulation.However, cur-rents passed through multiple electrodes simultaneously often combine nonlinearly to drive RGCresponses (nonlinear responses), making the responses difficult to predict and control. Additionally,multi-electrode stimulation is infeasible today because exhaustive calibration efforts in implanteddevices can grow exponentially with the number of electrodes. To overcome these challenges, this work leverages biophysical modeling of RGCs to better un-derstand nonlinear responses to multi-electrode stimulation and to aid the design of novel multi-electrode stimulation strategies. First, I present a biophysical model of electrical activation of a single RGC, developed using activemembrane properties from previous work and calibrated with electrical stimulation and recordingdata from a high-density microelectrode array (MEA). The model reproduces key trends, includingsigmoidal RGC response as a function of current with activation thresholds of 1-4μA, distinct bipha-sic and triphasic voltage waveforms of recorded spikes at different RGC compartments, and realisticspatial distribution of recorded spiking activity across the MEA, achieving an average correlationcoefficient of 0.82 between empirical and modeled data. Second, multi-electrode stimulation using two or three electrodes is modeled. The model correctlyreproduces nonlinear RGC responses to multi-electrode stimulation as well as cases where currents from multiple electrodes combine linearly to drive RGC responses (linear responses). We show thatthe distance and orientation of electrodes relative to a target RGC predict the linearity of resultingresponses, and these findings are consistent with trends seen in experimental data. Finally, these modeling insights guide the design of multi-electrode stimulation strategies thatexploit linear responses and avoid nonlinear responses.One such strategy, bi-electrode stimula-tion for axon avoidance, is tested experimentally and demonstrates, on average, a⇠21% reductionin target RGC thresholds and⇠14% increase in thresholds of non-target passing axons, a 1.44⇥selectivity improvement.Additional simulations project up to 3⇥selectivity improvements withmulti-electrode strategies compared to traditional single-electrode approaches. Acknowledgments "It takes a village" is an old adage that certainly rings true when describing my Ph.D., which wouldnot have been possible without the support of so many colleagues, friends, and family. I would like to thank my thesis co-advisors, Professors Subhasish Mitra and E.J. Chichilnisky,for their support and guidance. Subhasish challenged me to adopt a multi-disciplinary approach andtackle neuroscience problems with an engineer’s mindset.He consistently encouraged me to divedeeper into issues while ensuring I never lost sight of the overarching vision. E.J. is among the mostpassionate and dedicated scientists I have encountered during my PhD journey.His unwaveringdedication to the important