study, model & interface with motor cortex presented by - waseem khatri
TRANSCRIPT
STUDY, MODEL & INTERFACE WITH MOTOR CORTEX
Presented by - Waseem Khatri
Objective : Efficient decoding of neural information
for implementating neural motor prostheses
Motivation: Variety of Applications of Neural
Protheses Amputees can use artificial limbs Patients with Parkinsons disease Patients with paralysis/ spinal cord
injuries Epileptic seizures can be controlled
Brain Anatomy
Regions of the Brain What controls the Motor Skills ? Discovery ! Where is the Motor cortex located ?
Area 6 is further divided into-Pre-motor area – Motion Control
- Supplementary motor area – Motion Planning
Source: Macgill University
Homunculus
Little Man Somatotopic Representation Finest movements take more space Lips, Hands, Face have large areas in motor cortex
Source: http://thebrain.mcgill.ca/flash/a/a_06/a_06_cr/a_06_cr_mou/a_06_cr_mou.html
Source: Mijail Surruya
Experimental Setup for Neural Prostheses
Brain Controlled Vehicle for Paraplegic
Neural Interface Neural Signals
Sensors
Control Command
Vehicle State Signal
Vehicle
Environmental Feedback
Directional control
Key Questions ?
Measurements -What can we measure? -From where ? -How ?
Encoding– How is the information represented in the
brain?
Decoding – What algorithms can we use to infer the
internal state of the brain ? Interface
- Measurement
Source: Brown University
Why Primary Motor Cortex or M1 region ?• Firing rates of cells
correlated with hand motion (velocity, position, acceleration)
• Easily accessible
• Natural choice for controlling motion of a prosthetic device
- Encoding TechniquesSome of the encoding techniques used are
Population Vector- Neurons in M1 are broadly tuned to the direction of hand
movement, with each neuron having a preferred direction of movement for which its firing rate is maximal
Linear Filtering- Cells in M1 encode muscle activity in a linear fashion
Artificial Neural Networks
The Problem ?
Each of these methods estimates the hand kinematics x as a function of neural firing z
Encoding methods consider neural firing z as a function of hand kinematics x + noise.
But hand kinematics like – position, direction, velocity, acceleration etc. are considered in isolation
Hence , not very accurate results !Solution : Bayesian Population decoding using
Kalman filter
The Experiment
•Pursuit Tracking Task
•Pinball Task
•Record the Neural Activity
•Record the Hand Kinematics
•Compute the posterior probability of the hand motion conditioned on a sequence of observed firing rates; this is formulated in terms of the product of likelihood and a prior
•The likelihood term models the probability of firing rates given a particular hand motion and can be learned from training data.
•The prior term defines a probabilistic model of hand kinematics
Why Kalman Filter ?
Both the models : Likelihood and Prior are considered to be Gaussian.
A Kalman filter provides an efficient recursive method for Bayesian inference or estimating the posterior probability for the given two assumptions
Pin Ball Task
Source: Brown University
Generative Decoding Model
Source: Brown University
H is a matrix that linearly relates the six-dimensional hand state to the firing rates
A is the coefficient matrix
Markov Assumptions
given the hand kinematics at time k-1, the hand kinematics at time k is conditionally independent of the previous hand motions
)/(),.....,/( 112,1 kkkkk xxpxxxxp
)/(),/( 1 kkkkk xzpzxzp
conditioned on the current state, the firing rates are independent of the firing rates at previous time instants
Bayesian Inference
Source: Brown University
The posterior probability of the hand motion conditioned on a sequence of observed firing rates = The product of likelihood and a prior
Decoding involves estimating the posterior probability at each time instant
Bayesian Decision based Classifier
To simulate decision process after decoding
Prosthetic Arm Motion Classes – Flex and Extention 2 sets of Neuron outputs Training Data Assumption: Gaussian Model
Probability Density Functions
Class PDF’s
Classifier
Thank You
Questions ?