Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
AI-based optimization of somatosensory neuroprosthetic stimulation through learning control algorithms
The somatosensory nerves organization widely varies from one individual to another and even over time for the same subject. We propose that learning algorithms could be used to quickly optimize the stimulation parameters to be able to rapidly adapt these parameters according to the changes occuring.
Somatosensory neuroprostheses exploit invasive and non-invasive feedback technologies to restore sensorimotor functions lost to disease or trauma. These devices use electrical stimulation to communicate sensory information to the brain. A sensation characterization procedure is thus necessary to determine the appropriate stimulation parameters and to establish a clear personalized map of the sensations that can be restored. Several questionnaires have been described in the literature to collect the quality, type, location and intensity of the evoked sensations. The platform collects stimulation parameters used to elicit sensations; records subjects’ percepts in terms of sensation location, type, quality, perceptual threshold, and intensity. records subjects’ percepts in terms of sensation location, type, quality, perceptual threshold, and intensity.
Intraneural stimulation
The basic setup of a neuroprosthesis is as follows: data gathered by a sensor at the affected extremity, e.g. the sole of a foot prosthesis, is processed and transformed into stimulation patterns. According to these patterns, current pulses are generated and injected into the nerve via an implanted electrode to evoke action potentials within the targeted fibers The underlying methodology relies on the "hybrid electro-neuro" model, comprised of a module for Finite Element Method (FEM) electric field simulation and another for neural fibers signal propagation.Electro-Neuro model workflow
Available datasets for nerve geometry and structure are read into COMSOL (COMSOL Inc.). In this simulation environment, a three-dimensional FEM model with an electrical description of the nerve is created, with given parameters such as the injected current and flux at the active sites and impedance and conductivity throughout the nerve fascicle. Output of this simulation consists of the changes in electric field in the nervous tissue obtained from the intraneural stimulation. To complete the hybrid model, the result is processed by NEURON, another simulation environment. Neural fibers distribution inside the nerve fascicle is estimated and signal propagation induced by the evoked electrical field changes to provide a functional measure of the simulated neural interface.
Somatosensory neuroprostheses exploit invasive and non-invasive feedback technologies to restore sensorimotor functions lost to disease or trauma. These devices use electrical stimulation to communicate sensory information to the brain. A sensation characterization procedure is thus necessary to determine the appropriate stimulation parameters and to establish a clear personalized map of the sensations that can be restored. Several questionnaires have been described in the literature to collect the quality, type, location and intensity of the evoked sensations. The platform collects stimulation parameters used to elicit sensations; records subjects’ percepts in terms of sensation location, type, quality, perceptual threshold, and intensity. records subjects’ percepts in terms of sensation location, type, quality, perceptual threshold, and intensity. Intraneural stimulation The basic setup of a neuroprosthesis is as follows: data gathered by a sensor at the affected extremity, e.g. the sole of a foot prosthesis, is processed and transformed into stimulation patterns. According to these patterns, current pulses are generated and injected into the nerve via an implanted electrode to evoke action potentials within the targeted fibers The underlying methodology relies on the "hybrid electro-neuro" model, comprised of a module for Finite Element Method (FEM) electric field simulation and another for neural fibers signal propagation.Electro-Neuro model workflow Available datasets for nerve geometry and structure are read into COMSOL (COMSOL Inc.). In this simulation environment, a three-dimensional FEM model with an electrical description of the nerve is created, with given parameters such as the injected current and flux at the active sites and impedance and conductivity throughout the nerve fascicle. Output of this simulation consists of the changes in electric field in the nervous tissue obtained from the intraneural stimulation. To complete the hybrid model, the result is processed by NEURON, another simulation environment. Neural fibers distribution inside the nerve fascicle is estimated and signal propagation induced by the evoked electrical field changes to provide a functional measure of the simulated neural interface.
Here, we tackle the problem of controlling multi- dimensional neurostimulation intervention. This is a high-dimensional problem with very inefficient computational power use (e.g. the solving of singular problems is extremely slow). In practice, in intraneural stimulation of PNS as other functional electrical stimulation applications, stimulus parameters are often arbitrary and the strategies for optimizing these stimulations are poorly developed. Moreover, the somatosensory nerves organization widely varies from one individual to another and even over time for the same subject. We propose that learning algorithms could be used to quickly optimize the stimulation parameters of an implant according to the specific effects evoked by its electrodes and to be able to rapidly adapt these parameters according to the changes that may occur in the nerve. This would maximize the effectiveness and durability of somatosensory neuroprostheses.
We propose to use, as a starting point, Bayesian learning algorithm based on Gaussian processes to explore the space of stimulation parameters. We further want to to demonstrate that the algorithm can be applied to online optimization of somatosensory neuroprosthetic stimulation. The algorithm could find optimal nerve locations for stimulus delivery in less than half a minute and during active usage of the neuroprosthesis, largely outperforming the capacity of a human operator.
Here, we tackle the problem of controlling multi- dimensional neurostimulation intervention. This is a high-dimensional problem with very inefficient computational power use (e.g. the solving of singular problems is extremely slow). In practice, in intraneural stimulation of PNS as other functional electrical stimulation applications, stimulus parameters are often arbitrary and the strategies for optimizing these stimulations are poorly developed. Moreover, the somatosensory nerves organization widely varies from one individual to another and even over time for the same subject. We propose that learning algorithms could be used to quickly optimize the stimulation parameters of an implant according to the specific effects evoked by its electrodes and to be able to rapidly adapt these parameters according to the changes that may occur in the nerve. This would maximize the effectiveness and durability of somatosensory neuroprostheses. We propose to use, as a starting point, Bayesian learning algorithm based on Gaussian processes to explore the space of stimulation parameters. We further want to to demonstrate that the algorithm can be applied to online optimization of somatosensory neuroprosthetic stimulation. The algorithm could find optimal nerve locations for stimulus delivery in less than half a minute and during active usage of the neuroprosthesis, largely outperforming the capacity of a human operator.
Dr. Stanisa Raspopovic, Assistant Professor Neuroeengineering laboratory, Head
ETH Zurich, Switzerland
Email: stanisa.raspopovic@hest.ethz.ch
Dr. Giacomo Valle, Postdoc Neuroeengineering laboratory, ETH Zurich, Switzerland
Email: vallegiacomo@gmail.com
Dr. Stanisa Raspopovic, Assistant Professor Neuroeengineering laboratory, Head ETH Zurich, Switzerland Email: stanisa.raspopovic@hest.ethz.ch Dr. Giacomo Valle, Postdoc Neuroeengineering laboratory, ETH Zurich, Switzerland Email: vallegiacomo@gmail.com