Figure 1
From: Leveraging neural dynamics to extend functional lifetime of brain-machine interfaces
Decoder hysteresis and exploiting neural dynamical invariance. (a) Illustration of the decoder hysteresis concept. The blue curve represents the hypothetical drop off in performance in the scenario where neural recording electrodes are lost and a decoder is retrained naïvely with the remaining neurons. The red curve illustrates the idea of using prior information regarding motor cortex, specifically knowledge about neural dynamics, to augment the decode algorithm and mitigate the loss of performance as neurons are lost. (b) Block diagram for retraining with remaining neural signals. The present neurons are used to retrain a decode algorithm, which involves neural systems identification and decoder parameter learning to predict kinematics. When decoding withheld testing data to reconstruct a monkey’s hand velocity in the horizontal direction (gray trace), the retrained decoder (blue trace) poorly reconstructs the hand velocities. (c) Block diagram for decoder hysteresis. A historical dataset is used to identify neural dynamics of motor cortical activity. These dynamics are remembered and constrain the learning of decoder parameters with the present neural signals. With this approach, the decoded velocity (red) better reconstructs the hand velocities.
