Transcript Document
Cortical Neural Prosthetics Presented by Artie Wu Cortical Neural Prosthetics • Andrew B. Schwartz – Department of Neurobiology and Bioengineering, University of Pittsburgh – Annual Review of Neuroscience. 2004. 27:487-507 Outline • Background on Neural Prosthetics • Motivation • Electrodes – Microwires – ‘Michigan’ probes – Cyberkinetics probes • Complications • Extraction Algorithms • FLAMES: Floating Light Activated Micro Electrical Stimulator Neural Prostheses • Stimulating and recording electrodes implanted in cerebral cortex to activate neurons in different parts of CNS • Cortical Neural Prostheses (CNP) to control arm movement – Use neural activity to control devices to replace natural, animate movements in paralyzed individuals Current Benefits of Neural Prostheses • • • • • Restore hearing, vision Alleviate symptoms of Parkinson’s Disease (PD) Rid Tourette syndrome Mitigate head or spinal-cord trauma Restore movement in paralyzed patients Ridding Tourette Source: University Hospitals of Cleveland, affiliated with CWRU Problems of Muscle Activation • Muscle activation muscle force is nonlinear problem • Primary motor cortex drives motor activation – Depends on force, muscle length, limb geometry, orientation of limb relative to external forces, and inertia of moving segments Representing Movement • Current CNPs represent movement as end point displacements – Motor cortical discharge rate proportional to tuning function (discharge rate related to direction) • All cells actively code each direction • Weighted response gives specific direction using Population Vector Algorithm (PVA) • Magnitude and direction of this neural vector representation is highly correlated with movement velocity CNP: 3 Components • Record neural activity – Microelectrodes and recording electrodes • Extraction algorithm of neural code – Real time data acquisition and conversion to end point positions • Actuators – Animated computer displays, movement of robot arm, or activation of muscle Electrodes: Microwires • First chronic recording electrodes • 20-50μm diameter • Optimal insertion depth uncertain Electrodes: Silicon Micromachined Microprobes • ‘Michigan’ probe – Planar devices – Boron diffusion delineates shape of probe – Multiple recording sites along shaft Electrodes: Silicon Micromachined Microprobes • Cyberkinetics Inc/University of Utah array – 100 microelectrodes array on 4mm by 4mm base – Array inserted into cortex Tissue Reactions • Blood vessels disrupted and microhemorrhage • Astrocytes proliferate and form encapsulation around electrode • Cellular sheath around electrode with dense cells • Swelling pushes neurons away • Neuron density is increases after several weeks Impedance Caused by Encapsulation Source: ‘Chronic neural recordings using silicon microelectrode arrays electrochemically deposited with a poly(3,4-ethylenedioxythiophene) (PEDOT) film’, K. Ludwig, J. Neural Eng. 3. 2006, 59-70. Extraction algorithms: Inferential • Population Vector Algorithm relates movement direction and firing rate in motor cortex D – bo = A ·cosθ = bxmx + bymy + bzmz D: discharge rate A: amplitude of tuning function Θ: angle between cell’s preferred direction B: vector in direction of preferred direction, magnitude is A M: unit vector in movement direction Extraction algorithms: Classifiers • Based on pattern recognition • Ex: Self-organizing feature map (SOFM) – Single layer of nodes connected to input vector with set of connection weight (i.e., discharge rate) – Initially, weight vectors set randomly – Element with weight vector closest to input vector = winner – Neighbor’s weights moved closer to input vectors – Each cluster assigned direction FLAMES: Floating Light Activated Micro Electrical Stimulators Source: Steve Menn FLAMES: Floating Light Activated Micro Electrical Stimulators Source: Steve Menn FLAMES: Floating Light Activated Micro Electrical Stimulators Source: Steve Menn Round 3 Design Specs • 56 different devices • 1,2,3,4 diodes • With and without 50kΩ parallel resistor Thank you