Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
, 23 (12), 3162-204

Adaptive Decoding for Brain-Machine Interfaces Through Bayesian Parameter Updates

Affiliations

Adaptive Decoding for Brain-Machine Interfaces Through Bayesian Parameter Updates

Zheng Li et al. Neural Comput.

Abstract

Brain-machine interfaces (BMIs) transform the activity of neurons recorded in motor areas of the brain into movements of external actuators. Representation of movements by neuronal populations varies over time, during both voluntary limb movements and movements controlled through BMIs, due to motor learning, neuronal plasticity, and instability in recordings. To ensure accurate BMI performance over long time spans, BMI decoders must adapt to these changes. We propose the Bayesian regression self-training method for updating the parameters of an unscented Kalman filter decoder. This novel paradigm uses the decoder's output to periodically update its neuronal tuning model in a Bayesian linear regression. We use two previously known statistical formulations of Bayesian linear regression: a joint formulation, which allows fast and exact inference, and a factorized formulation, which allows the addition and temporary omission of neurons from updates but requires approximate variational inference. To evaluate these methods, we performed offline reconstructions and closed-loop experiments with rhesus monkeys implanted cortically with microwire electrodes. Offline reconstructions used data recorded in areas M1, S1, PMd, SMA, and PP of three monkeys while they controlled a cursor using a handheld joystick. The Bayesian regression self-training updates significantly improved the accuracy of offline reconstructions compared to the same decoder without updates. We performed 11 sessions of real-time, closed-loop experiments with a monkey implanted in areas M1 and S1. These sessions spanned 29 days. The monkey controlled the cursor using the decoder with and without updates. The updates maintained control accuracy and did not require information about monkey hand movements, assumptions about desired movements, or knowledge of the intended movement goals as training signals. These results indicate that Bayesian regression self-training can maintain BMI control accuracy over long periods, making clinical neuroprosthetics more viable.

Figures

Figure 1
Figure 1
Implants and experimental methods. A: Implant placement and geometries. Micro-wire diameters are specified in microns. To reduce clutter, we only show arrays from which data were used. B: Primate experiment setup. C: Tuning model update paradigm. Updates occur periodically, using the neuronal data and decoder outputs generated since the previous update. D: Three behavioral tasks. Each task required placing the cursor over a target. In center-out, targets were stationary. In the pursuit tasks, targets moved continuously according to a Lissajous curve or a set of invisible, random waypoints.
Figure 2
Figure 2
Graphical models for A. joint distribution Bayesian regression, B. factorized distribution variational Bayesian regression. Ellipses indicate random variables, double ellipses indicate observed random variables, rectangles indicate observed point values, and rectangles indicate repeated nodes.
Figure 3
Figure 3
Position reconstruction accuracy of UKF with and without updates. A. Bars summarize mean accuracy for each monkey and condition. The static condition used parameters fit with joint distribution Bayesian regression. BR-hand and VBR-hand conditions used hand movements for updates. Accuracy values are from the best setting of the model drift parameter found using the test data. B. Accuracy improvement versus model drift parameter. Sessions indicated by curve labels.
Figure 4
Figure 4
Closed-loop BMI control accuracy with and without updates. A. Solid curves indicate SNR over the entire condition time segment of each session, and dotted curves indicate the best SNR among 1-minute windows in the condition time segment of each session. In the last two sessions, we updated the baseline firing rate parameters of the static UKF using exponential moving averages. B. Closed-loop BMI control accuracy versus time within each session. In four sessions, control accuracy was poor initially, but improved after two to three updates.
Figure 5
Figure 5
Traces of closed-loop BMI control with the UKF updated with VBR and with a static UKF. Plots show x-axis position, dark curves indicate the target location, and light curves indicate the BMI controlled cursor location.
Figure 6
Figure 6
Changes in tuning parameters detected by the adaptive decoder during closed-loop sessions. A. Histogram of absolute changes in tuning model coefficients (H). B. Histogram of changes in coefficient precisions (diagonals of Λi). C. Histogram of changes in baseline firing rates. D. Histogram of changes in noise standard deviation (square root of diagonal of R).

Similar articles

See all similar articles

Cited by 34 PubMed Central articles

See all "Cited by" articles

Publication types

LinkOut - more resources

Feedback