Limits...
Implications of the Dependence of Neuronal Activity on Neural Network States for the Design of Brain-Machine Interfaces.

Panzeri S, Safaai H, De Feo V, Vato A - Front Neurosci (2016)

Bottom Line: Brain-machine interfaces (BMIs) can improve the quality of life of patients with sensory and motor disabilities by both decoding motor intentions expressed by neural activity, and by encoding artificially sensed information into patterns of neural activity elicited by causal interventions on the neural tissue.Yet, current BMIs can exchange relatively small amounts of information with the brain.We then elaborate on how this knowledge may be used to increase the amount of information that BMIs exchange with brain.

View Article: PubMed Central - PubMed

Affiliation: Neural Computation Laboratory, Istituto Italiano di Tecnologia Rovereto, Italy.

ABSTRACT
Brain-machine interfaces (BMIs) can improve the quality of life of patients with sensory and motor disabilities by both decoding motor intentions expressed by neural activity, and by encoding artificially sensed information into patterns of neural activity elicited by causal interventions on the neural tissue. Yet, current BMIs can exchange relatively small amounts of information with the brain. This problem has proved difficult to overcome by simply increasing the number of recording or stimulating electrodes, because trial-to-trial variability of neural activity partly arises from intrinsic factors (collectively known as the network state) that include ongoing spontaneous activity and neuromodulation, and so is shared among neurons. Here we review recent progress in characterizing the state dependence of neural responses, and in particular of how neural responses depend on endogenous slow fluctuations of network excitability. We then elaborate on how this knowledge may be used to increase the amount of information that BMIs exchange with brain. Knowledge of network state can be used to fine-tune the stimulation pattern that should reliably elicit a target neural response used to encode information in the brain, and to discount part of the trial-by-trial variability of neural responses, so that they can be decoded more accurately.

No MeSH data available.


Related in: MedlinePlus

How knowledge of state dependence may be used to improve reliability of elicited patterns and to enhance decoding of neural activity. This figure uses cartoons of neural responses to illustrate how taking into account state dependence of neural responses may improve BMIs. (A) The panel illustrates the responses of a cartoon neuron to two different stimuli presented at different phases of LFP. Green and pink arrows represent the application times of stimulus s = 1 and s = 2, respectively. Stimuli applied at the trough of the LFP (a more excitable network state phase) elicit a stronger response than stimuli applied at the peak of the LFP (a less excitable network state phase). (B) We plot the time course of cartoon neural activity shortly after a stimulus is applied at time t = 0. Different lines represent single-trial responses to the same stimulus that were elicited in trials that differed by the value of the state variable (in this case, the LFP phase) in each trial. Each line representing the single-trial responses is color-coded by the value of the state variable in that trial. The response to the same stimulus has large trial-to-trial variability because of the difference in the state at which the stimulus is presented. (C) In a particular network state, the real response (dashed black line) and the state-dependent model prediction (pink line) of the response are shown. If the model of state dependence is accurate, it will help the experimenter to predict which response will be elicited in that trial given the stimulation parameters and the neural state, thereby narrowing down the uncertainty about which response will be elicited. The gray area shows the range of possible responses that could be obtained for a given stimulus because of state variations. (D) A scatter plot showing the variations around the mean of single-trial responses plotted against the single-trial response variations around the mean estimated from a state-dependent model of the responses in a hypothetical case. Each point represents one trial. This scatter plot indicates how well a state dependence model can predict the single-trial neural responses to a given stimulus. (E) How to discount state-induced trial-to-trial variability is exemplified for a single trial. Stimulus 1 was presented in this cartoon trial, and the response in this trial (full black line) is plotted against the trial-averaged response of stimulus 1 (green line) and of stimulus 2 (pink line). The state dependence model predicts that the response variation around the trial-average in this trial was negative. The black arrows show the model-predicted state-induced variation around the trial-average for this particular single trial. The addition of the model predicted variability gives a “discounted” response (dashed black line) that will be much closer to the averaged response to stimulus 1 (the stimulus actually presented in this trial) that the original response. (F) The distributions of the responses to stimulus 1 (green full line) and stimulus 2 (pink full line) and the distributions of the discounted responses for stimulus 1 (dashed green line) and stimulus 2 (dashed pink line) obtained after subtracting the model predicted state-induced variability are shown. The distributions of the discounted responses are narrower and allow better discrimination between the two stimuli.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4837323&req=5

Figure 2: How knowledge of state dependence may be used to improve reliability of elicited patterns and to enhance decoding of neural activity. This figure uses cartoons of neural responses to illustrate how taking into account state dependence of neural responses may improve BMIs. (A) The panel illustrates the responses of a cartoon neuron to two different stimuli presented at different phases of LFP. Green and pink arrows represent the application times of stimulus s = 1 and s = 2, respectively. Stimuli applied at the trough of the LFP (a more excitable network state phase) elicit a stronger response than stimuli applied at the peak of the LFP (a less excitable network state phase). (B) We plot the time course of cartoon neural activity shortly after a stimulus is applied at time t = 0. Different lines represent single-trial responses to the same stimulus that were elicited in trials that differed by the value of the state variable (in this case, the LFP phase) in each trial. Each line representing the single-trial responses is color-coded by the value of the state variable in that trial. The response to the same stimulus has large trial-to-trial variability because of the difference in the state at which the stimulus is presented. (C) In a particular network state, the real response (dashed black line) and the state-dependent model prediction (pink line) of the response are shown. If the model of state dependence is accurate, it will help the experimenter to predict which response will be elicited in that trial given the stimulation parameters and the neural state, thereby narrowing down the uncertainty about which response will be elicited. The gray area shows the range of possible responses that could be obtained for a given stimulus because of state variations. (D) A scatter plot showing the variations around the mean of single-trial responses plotted against the single-trial response variations around the mean estimated from a state-dependent model of the responses in a hypothetical case. Each point represents one trial. This scatter plot indicates how well a state dependence model can predict the single-trial neural responses to a given stimulus. (E) How to discount state-induced trial-to-trial variability is exemplified for a single trial. Stimulus 1 was presented in this cartoon trial, and the response in this trial (full black line) is plotted against the trial-averaged response of stimulus 1 (green line) and of stimulus 2 (pink line). The state dependence model predicts that the response variation around the trial-average in this trial was negative. The black arrows show the model-predicted state-induced variation around the trial-average for this particular single trial. The addition of the model predicted variability gives a “discounted” response (dashed black line) that will be much closer to the averaged response to stimulus 1 (the stimulus actually presented in this trial) that the original response. (F) The distributions of the responses to stimulus 1 (green full line) and stimulus 2 (pink full line) and the distributions of the discounted responses for stimulus 1 (dashed green line) and stimulus 2 (dashed pink line) obtained after subtracting the model predicted state-induced variability are shown. The distributions of the discounted responses are narrower and allow better discrimination between the two stimuli.

Mentions: To discuss this, we will suppose for simplicity that, as in auditory cortex (Kayser et al., 2015), the response r of a neuron to a stimulus s in a single trial tr depends on a state variable θ (in this example, the phase of a low-frequency LFP at which stimulus s is applied in trial tr) with an additive-multiplicative model. The gain g and the background b both depend on the phase of a low-frequency network oscillation at which the stimulus is applied(1)r(s,tr)=g(θ)f(s)+b(θ)where state θ is function of the trial and f(s) is the trial-averaged responses to all trials to stimulus s [in other words, f(s) is the neuron tuning's curve]. We assume that, again as in Kayser et al. (2015), if the stimulus was presented in the phases of the low-frequency LFP that corresponded to LFP troughs (respectively, peaks) then it elicited a larger (respectively, smaller) response because both the gain and the background activity were larger (respectively, smaller), see Figure 2A.


Implications of the Dependence of Neuronal Activity on Neural Network States for the Design of Brain-Machine Interfaces.

Panzeri S, Safaai H, De Feo V, Vato A - Front Neurosci (2016)

How knowledge of state dependence may be used to improve reliability of elicited patterns and to enhance decoding of neural activity. This figure uses cartoons of neural responses to illustrate how taking into account state dependence of neural responses may improve BMIs. (A) The panel illustrates the responses of a cartoon neuron to two different stimuli presented at different phases of LFP. Green and pink arrows represent the application times of stimulus s = 1 and s = 2, respectively. Stimuli applied at the trough of the LFP (a more excitable network state phase) elicit a stronger response than stimuli applied at the peak of the LFP (a less excitable network state phase). (B) We plot the time course of cartoon neural activity shortly after a stimulus is applied at time t = 0. Different lines represent single-trial responses to the same stimulus that were elicited in trials that differed by the value of the state variable (in this case, the LFP phase) in each trial. Each line representing the single-trial responses is color-coded by the value of the state variable in that trial. The response to the same stimulus has large trial-to-trial variability because of the difference in the state at which the stimulus is presented. (C) In a particular network state, the real response (dashed black line) and the state-dependent model prediction (pink line) of the response are shown. If the model of state dependence is accurate, it will help the experimenter to predict which response will be elicited in that trial given the stimulation parameters and the neural state, thereby narrowing down the uncertainty about which response will be elicited. The gray area shows the range of possible responses that could be obtained for a given stimulus because of state variations. (D) A scatter plot showing the variations around the mean of single-trial responses plotted against the single-trial response variations around the mean estimated from a state-dependent model of the responses in a hypothetical case. Each point represents one trial. This scatter plot indicates how well a state dependence model can predict the single-trial neural responses to a given stimulus. (E) How to discount state-induced trial-to-trial variability is exemplified for a single trial. Stimulus 1 was presented in this cartoon trial, and the response in this trial (full black line) is plotted against the trial-averaged response of stimulus 1 (green line) and of stimulus 2 (pink line). The state dependence model predicts that the response variation around the trial-average in this trial was negative. The black arrows show the model-predicted state-induced variation around the trial-average for this particular single trial. The addition of the model predicted variability gives a “discounted” response (dashed black line) that will be much closer to the averaged response to stimulus 1 (the stimulus actually presented in this trial) that the original response. (F) The distributions of the responses to stimulus 1 (green full line) and stimulus 2 (pink full line) and the distributions of the discounted responses for stimulus 1 (dashed green line) and stimulus 2 (dashed pink line) obtained after subtracting the model predicted state-induced variability are shown. The distributions of the discounted responses are narrower and allow better discrimination between the two stimuli.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4837323&req=5

Figure 2: How knowledge of state dependence may be used to improve reliability of elicited patterns and to enhance decoding of neural activity. This figure uses cartoons of neural responses to illustrate how taking into account state dependence of neural responses may improve BMIs. (A) The panel illustrates the responses of a cartoon neuron to two different stimuli presented at different phases of LFP. Green and pink arrows represent the application times of stimulus s = 1 and s = 2, respectively. Stimuli applied at the trough of the LFP (a more excitable network state phase) elicit a stronger response than stimuli applied at the peak of the LFP (a less excitable network state phase). (B) We plot the time course of cartoon neural activity shortly after a stimulus is applied at time t = 0. Different lines represent single-trial responses to the same stimulus that were elicited in trials that differed by the value of the state variable (in this case, the LFP phase) in each trial. Each line representing the single-trial responses is color-coded by the value of the state variable in that trial. The response to the same stimulus has large trial-to-trial variability because of the difference in the state at which the stimulus is presented. (C) In a particular network state, the real response (dashed black line) and the state-dependent model prediction (pink line) of the response are shown. If the model of state dependence is accurate, it will help the experimenter to predict which response will be elicited in that trial given the stimulation parameters and the neural state, thereby narrowing down the uncertainty about which response will be elicited. The gray area shows the range of possible responses that could be obtained for a given stimulus because of state variations. (D) A scatter plot showing the variations around the mean of single-trial responses plotted against the single-trial response variations around the mean estimated from a state-dependent model of the responses in a hypothetical case. Each point represents one trial. This scatter plot indicates how well a state dependence model can predict the single-trial neural responses to a given stimulus. (E) How to discount state-induced trial-to-trial variability is exemplified for a single trial. Stimulus 1 was presented in this cartoon trial, and the response in this trial (full black line) is plotted against the trial-averaged response of stimulus 1 (green line) and of stimulus 2 (pink line). The state dependence model predicts that the response variation around the trial-average in this trial was negative. The black arrows show the model-predicted state-induced variation around the trial-average for this particular single trial. The addition of the model predicted variability gives a “discounted” response (dashed black line) that will be much closer to the averaged response to stimulus 1 (the stimulus actually presented in this trial) that the original response. (F) The distributions of the responses to stimulus 1 (green full line) and stimulus 2 (pink full line) and the distributions of the discounted responses for stimulus 1 (dashed green line) and stimulus 2 (dashed pink line) obtained after subtracting the model predicted state-induced variability are shown. The distributions of the discounted responses are narrower and allow better discrimination between the two stimuli.
Mentions: To discuss this, we will suppose for simplicity that, as in auditory cortex (Kayser et al., 2015), the response r of a neuron to a stimulus s in a single trial tr depends on a state variable θ (in this example, the phase of a low-frequency LFP at which stimulus s is applied in trial tr) with an additive-multiplicative model. The gain g and the background b both depend on the phase of a low-frequency network oscillation at which the stimulus is applied(1)r(s,tr)=g(θ)f(s)+b(θ)where state θ is function of the trial and f(s) is the trial-averaged responses to all trials to stimulus s [in other words, f(s) is the neuron tuning's curve]. We assume that, again as in Kayser et al. (2015), if the stimulus was presented in the phases of the low-frequency LFP that corresponded to LFP troughs (respectively, peaks) then it elicited a larger (respectively, smaller) response because both the gain and the background activity were larger (respectively, smaller), see Figure 2A.

Bottom Line: Brain-machine interfaces (BMIs) can improve the quality of life of patients with sensory and motor disabilities by both decoding motor intentions expressed by neural activity, and by encoding artificially sensed information into patterns of neural activity elicited by causal interventions on the neural tissue.Yet, current BMIs can exchange relatively small amounts of information with the brain.We then elaborate on how this knowledge may be used to increase the amount of information that BMIs exchange with brain.

View Article: PubMed Central - PubMed

Affiliation: Neural Computation Laboratory, Istituto Italiano di Tecnologia Rovereto, Italy.

ABSTRACT
Brain-machine interfaces (BMIs) can improve the quality of life of patients with sensory and motor disabilities by both decoding motor intentions expressed by neural activity, and by encoding artificially sensed information into patterns of neural activity elicited by causal interventions on the neural tissue. Yet, current BMIs can exchange relatively small amounts of information with the brain. This problem has proved difficult to overcome by simply increasing the number of recording or stimulating electrodes, because trial-to-trial variability of neural activity partly arises from intrinsic factors (collectively known as the network state) that include ongoing spontaneous activity and neuromodulation, and so is shared among neurons. Here we review recent progress in characterizing the state dependence of neural responses, and in particular of how neural responses depend on endogenous slow fluctuations of network excitability. We then elaborate on how this knowledge may be used to increase the amount of information that BMIs exchange with brain. Knowledge of network state can be used to fine-tune the stimulation pattern that should reliably elicit a target neural response used to encode information in the brain, and to discount part of the trial-by-trial variability of neural responses, so that they can be decoded more accurately.

No MeSH data available.


Related in: MedlinePlus