Limits...
Exploiting Task Constraints for Self-Calibrated Brain-Machine Interface Control Using Error-Related Potentials.

Iturrate I, Grizou J, Omedes J, Oudeyer PY, Lopes M, Montesano L - PLoS ONE (2015)

Bottom Line: This paper presents a new approach for self-calibration BCI for reaching tasks using error-related potentials.The results show that it is possible to have a usable BCI control from the beginning of the experiment without any prior calibration.Furthermore, comparisons with simulations and previous results obtained using standard calibration hint that both the quality of recorded signals and the performance of the system were comparable to those obtained with a standard calibration approach.

View Article: PubMed Central - PubMed

Affiliation: Chair in Brain-Machine Interface (CNBI) and Center for Neuroprosthetics (CNP), École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland; Instituto de Investigación en Ingeniería de Sistemas (I3A), Universidad de Zaragoza, Zaragoza, Spain.

ABSTRACT
This paper presents a new approach for self-calibration BCI for reaching tasks using error-related potentials. The proposed method exploits task constraints to simultaneously calibrate the decoder and control the device, by using a robust likelihood function and an ad-hoc planner to cope with the large uncertainty resulting from the unknown task and decoder. The method has been evaluated in closed-loop online experiments with 8 users using a previously proposed BCI protocol for reaching tasks over a grid. The results show that it is possible to have a usable BCI control from the beginning of the experiment without any prior calibration. Furthermore, comparisons with simulations and previous results obtained using standard calibration hint that both the quality of recorded signals and the performance of the system were comparable to those obtained with a standard calibration approach.

No MeSH data available.


Percentage of errors performed by the device during the last 10 trials, as a function of the number of trials (only during the first target).Additionally, the tendency line and the correlation value are also shown.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4488878&req=5

pone.0131491.g003: Percentage of errors performed by the device during the last 10 trials, as a function of the number of trials (only during the first target).Additionally, the tendency line and the correlation value are also shown.

Mentions: Fig 3 shows the evolution of the ratio of errors within the last ten movements of the device until reaching the first target. The results for each subject show a clear decreasing trend in the percentage of errors significant for all the subjects (p < 0.001) with an average correlation of r = 0.57±0.17. Expectedly, these results show that the error rate was variable throughout the experiment. Nonetheless, the negative tendency also indicated that the system was able to iteratively learn the task even for the first target using completely unsupervised data. Regarding the labels quality, Fig 4a shows the ten-fold accuracy obtained with the ground truth labels, compared against the percentage of labels correctly learned by the self-calibration protocol. The results show an almost significant correlation between the two variables (r = 0.69, p = 0.06). Nonetheless, the ratio of labels correctly assigned was always higher than 90%, indicating that even for subjects with low accuracies (< 60%), most of the labels are correctly estimated. Fig 4b shows the ten-fold accuracy obtained with ground truth labels versus the accuracy obtained with the labels estimated from self-calibration. Interestingly, there seemed to be two clusters of subjects: those who were not affected by the labeling quality composed of five subjects with very similar accuracies to the ones obtained with ground truth labels, even when not all the labels were correctly assigned; and other subjects where the labeling of the self-calibration approach affected the separability of the data acquired with a decrease of 8.41% of accuracy on average.


Exploiting Task Constraints for Self-Calibrated Brain-Machine Interface Control Using Error-Related Potentials.

Iturrate I, Grizou J, Omedes J, Oudeyer PY, Lopes M, Montesano L - PLoS ONE (2015)

Percentage of errors performed by the device during the last 10 trials, as a function of the number of trials (only during the first target).Additionally, the tendency line and the correlation value are also shown.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4488878&req=5

pone.0131491.g003: Percentage of errors performed by the device during the last 10 trials, as a function of the number of trials (only during the first target).Additionally, the tendency line and the correlation value are also shown.
Mentions: Fig 3 shows the evolution of the ratio of errors within the last ten movements of the device until reaching the first target. The results for each subject show a clear decreasing trend in the percentage of errors significant for all the subjects (p < 0.001) with an average correlation of r = 0.57±0.17. Expectedly, these results show that the error rate was variable throughout the experiment. Nonetheless, the negative tendency also indicated that the system was able to iteratively learn the task even for the first target using completely unsupervised data. Regarding the labels quality, Fig 4a shows the ten-fold accuracy obtained with the ground truth labels, compared against the percentage of labels correctly learned by the self-calibration protocol. The results show an almost significant correlation between the two variables (r = 0.69, p = 0.06). Nonetheless, the ratio of labels correctly assigned was always higher than 90%, indicating that even for subjects with low accuracies (< 60%), most of the labels are correctly estimated. Fig 4b shows the ten-fold accuracy obtained with ground truth labels versus the accuracy obtained with the labels estimated from self-calibration. Interestingly, there seemed to be two clusters of subjects: those who were not affected by the labeling quality composed of five subjects with very similar accuracies to the ones obtained with ground truth labels, even when not all the labels were correctly assigned; and other subjects where the labeling of the self-calibration approach affected the separability of the data acquired with a decrease of 8.41% of accuracy on average.

Bottom Line: This paper presents a new approach for self-calibration BCI for reaching tasks using error-related potentials.The results show that it is possible to have a usable BCI control from the beginning of the experiment without any prior calibration.Furthermore, comparisons with simulations and previous results obtained using standard calibration hint that both the quality of recorded signals and the performance of the system were comparable to those obtained with a standard calibration approach.

View Article: PubMed Central - PubMed

Affiliation: Chair in Brain-Machine Interface (CNBI) and Center for Neuroprosthetics (CNP), École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland; Instituto de Investigación en Ingeniería de Sistemas (I3A), Universidad de Zaragoza, Zaragoza, Spain.

ABSTRACT
This paper presents a new approach for self-calibration BCI for reaching tasks using error-related potentials. The proposed method exploits task constraints to simultaneously calibrate the decoder and control the device, by using a robust likelihood function and an ad-hoc planner to cope with the large uncertainty resulting from the unknown task and decoder. The method has been evaluated in closed-loop online experiments with 8 users using a previously proposed BCI protocol for reaching tasks over a grid. The results show that it is possible to have a usable BCI control from the beginning of the experiment without any prior calibration. Furthermore, comparisons with simulations and previous results obtained using standard calibration hint that both the quality of recorded signals and the performance of the system were comparable to those obtained with a standard calibration approach.

No MeSH data available.