Biomimetic multichannel neurostimulation
Sensory information can be delivered to a subject mammal, for example, for restoring a sense of cutaneous touch and limb motion to the subject mammal. A biomimetic electrical signal is generated based on (a) a stimulation reference signal applied to a somatosensory region of a nervous system of a reference mammal, (b) a stimulated-response signal acquired from a sensory cortex of the reference mammal in response to application of the stimulation reference signal to the thalamic nucleus, and (c) a natural-response signal acquired from the sensory cortex in response to peripheral touch stimuli and/or peripheral nerve stimulation of the reference mammal. The biomimetic electrical signal is applied to a somatosensory region of a nervous system of the subject mammal to induce an activation response, in a sensory cortex of the subject mammal.
1. A method of delivering sensory information to a subject mammal, the method comprising:
generating a biomimetic electrical signal based on (a) a stimulation reference signal applied to a somatosensory region of a nervous system of a reference mammal, (b) a stimulated-response signal acquired from a sensory cortex of the reference mammal in response to application of the stimulation reference signal to the thalamic nucleus, and (c) a natural-response signal acquired from the sensory cortex in response to peripheral touch stimuli and/or peripheral nerve stimulation of the reference mammal; and
applying the biomimetic electrical signal to a somatosensory region of a nervous system of the subject mammal to induce an activation response.
2. The method of claim 1 , wherein the reference mammal is a mammal other than the subject mammal.
3. The method of claim 1 , wherein the sensory cortex of each of the reference mammal and the subject mammal is a primary somatosensory cortex (S1).
4. The method of claim 1 , wherein the somatosensory region of each of the reference mammal and the subject mammal comprises a primary somatosensory cortex.
5. The method of claim 4 , wherein the somatosensory region of each of the reference mammal and the subject mammal comprises at least one of a Brodmann area 1, a Brodmann area 2, a Brodmann area3b, a Brodmann area 3a and a somatosensory thalamic nucleus, such as a Ventral Posterior Lateral (VPL) nucleus of the thalamus, also called the Ventral Caudal (VC) nucleus in humans, or the proprioceptive thalamic region.
6. The method of claim 1 , wherein the somatosensory region of the nervous system of each of the reference mammal and the subject mammal comprises a thalamic nucleus.
7. The method of claim 6 , wherein the thalamic nucleus of each of the reference mammal and the subject mammal is a ventral posterior lateral thalamus, or Ventral Caudal (VC) nucleus.
8. The method of claim 1 , wherein the peripheral sensory nerve of the subject mammal is in a limb of the subject mammal.
9. The method of claim 8 , wherein the biomimetic electrical signal is applied to the thalamic nucleus of the subject mammal after the subject mammal has lost at least a portion of the limb.
10. The method of claim 1 , wherein the natural-response reference signal is acquired from the sensory cortex of the reference mammal in response to a mechanical stimulation of the peripheral sensory nerve of the reference mammal.
11. The method of claim 1 , wherein the applying of the biomimetic electrical signal is through multiple channels.
12. The method of claim 11 , wherein the applying is by a microelectrode array.
13. The method of claim 1 , wherein each of the biomimetic electrical signal, the stimulation reference signal, the stimulated-response signal, and the natural-response signal comprises a spatiotemporal pattern.
14. The method of claim 1 , wherein each of the stimulated-response signal and the natural-response signal comprises information regarding at least one of local field potentials, spike times/rates, or spike counts.
15. The method of claim 1 , wherein the biomimetic electrical signal is generated from an algorithm based on the stimulation reference signal, the stimulated-response signal, and the natural-response signal.
16. The method of claim 15 , wherein the algorithm is derived from a state-space model using the stimulation reference signal, the stimulated-response signal, and the natural-response signal.
17. The method of claim 1 , wherein the biomimetic electrical signal is generated by a model predictive controller using a state-space model trained from the stimulation reference signal and the stimulated-response signal and optimized using the natural-response signal.
18. The method of claim 17 , wherein the state-space model comprises a discrete-time linear dynamical model trained from the stimulation reference signal and the stimulated-response signal.
19. The method of claim 18 , wherein the discrete-time linear dynamical model is defined, at least in part, by equations (1) and (2):
x t+1 =Ax t +Bu t +ϵ x (1)
y t =Cx t +ϵ y (2)
wherein each vector containing stimulation channel magnitudes at time t is denoted u(t)ϵ m, the state is denoted by xϵ n, and the output is denoted yϵ p, ϵx˜N (0, Q) and ϵy˜N(0, R).
20. The method of claim 19 , wherein A, B, C, Q, and R are determined using system identification techniques including at least one of subspace identification, least-squares regression with filter bank features as inputs followed by a model reduction technique wherein A, B, and C represent local linearizations of a nonlinear model which include at least one of neural networks, Volterra series models, and kernel regression methods.
21. The method of claim 19 , wherein the model predictive controller minimizes a squared Euclidean distance between y(t) and a target output signal.
22. The method of claim 21 , wherein the model predictive controller operates to:
minimize z T Hz+κφ(z)
subject to Tz=0
wherein κ is a weighting parameter to prioritize the contribution of a log barrier, z is formed by stacking the states xt, ut for t=1, 2, . . . to the length of the control horizon, H represents quadratic penalties imposed by taking Euclidean distances between the output and a desired response signal for each time point, and T is a matrix representing the relationships between adjacent time points as in equation (1).
23. The method of claim 22 , wherein:
minimize z T Hz+κφ(z)
subject to Tz=0
is solved using convex optimization.
24. The method of claim 17 , wherein the biomimetic electrical signal is generated by the model predictive controller, and optimized using an average of a plurality of natural-response signals as target waveforms.
25. The method of claim 17 , wherein the biomimetic electrical signal is generated by the model predictive controller, optimizing using a target generated from a predictive neural encoding model of a plurality of natural-response signals.
26. The method of claim 17 , wherein the model predictive controller employs convex optimization.
27. The method of claim 17 , wherein the model predictive controller minimizes mean-square error.
28. The method of claim 1 , wherein the activation response comprises a signal emanating from the sensory cortex of the subject mammal.