![Incorporating decoder outputs in the MINDFUL score maintains a high correlation with performance
a The KLD (right y-axis) between distributions of decoded directional vectors, X̂\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{X}$$\end{document}, with respect to the sub-selected time bins from the first session(s) overlaid onto median AE (left y-axis in blue) across all recorded sessions for T11 (left panel) and T5 (right panel). Subsequent neural distributions and median AE were updated every 1 s over a 60-s sliding window. Pearson r, and Spearman rank correlation coefficients ρ, between KLD and median AE are shown. b The KLD between distributions of X̂\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{X}$$\end{document} and X̂lag\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{X}}_{{lag}}$$\end{document} overlaid onto median AE. c The KLD between distributions of the combination of derived neural features (as shown in Fig. 1c), decoded directional vectors, X̂\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{X}$$\end{document}, and X̂lag\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{X}}_{{lag}}$$\end{document}, overlaid onto median AE.](https://www.researchgate.net/publication/385107184/figure/fig2/AS:11431281285081991@1729566093814/Incorporating-decoder-outputs-in-the-MINDFUL-score-maintains-a-high-correlation-with_Q320.jpg)
![Sub-selecting instances of low AE for reference and using longer window length improve correlation between MINDFUL and AE
a Spearman correlation coefficients of KLD of NF, X̂\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{X}$$\end{document}, and X̂lag\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\hat{X}}_{lag}$$\end{document}, to AE when sub-selecting time bins with different quantiles of AE as the reference. Each quantile approximately contains an even number of observations. b Spearman correlation coefficients of KLD to AE when using different window lengths to estimate target distributions.](https://www.researchgate.net/publication/385107184/figure/fig5/AS:11431281284951263@1729566094752/Sub-selecting-instances-of-low-AE-for-reference-and-using-longer-window-length-improve_Q320.jpg)
October 2024
·
24 Reads
·
4 Citations
Communications Biology
Intracortical brain-computer interfaces (iBCIs) enable people with tetraplegia to gain intuitive cursor control from movement intentions. To translate to practical use, iBCIs should provide reliable performance for extended periods of time. However, performance begins to degrade as the relationship between kinematic intention and recorded neural activity shifts compared to when the decoder was initially trained. In addition to developing decoders to better handle long-term instability, identifying when to recalibrate will also optimize performance. We propose a method, “MINDFUL”, to measure instabilities in neural data for useful long-term iBCI, without needing labels of user intentions. Longitudinal data were analyzed from two BrainGate2 participants with tetraplegia as they used fixed decoders to control a computer cursor spanning 142 days and 28 days, respectively. We demonstrate a measure of instability that correlates with changes in closed-loop cursor performance solely based on the recorded neural activity (Pearson r = 0.93 and 0.72, respectively). This result suggests a strategy to infer online iBCI performance from neural data alone and to determine when recalibration should take place for practical long-term use.
