Journal: APL Bioengineering

Article Title: Degradation-aware neural imputation: Advancing decoding stability in brain machine interfaces

doi: 10.1063/5.0250296

Figure Lengend Snippet: Contribution ratio ( % Contribution ) of MUA and LFP features in predicting forelimb movement velocities along the x- and y-axes during the baseline period (recording days 1–7). (a) Contribution ratios of neural features for predicting forelimb movement velocity along the x-axis. (b) Contribution ratios of neural features for predicting forelimb movement velocity along the y-axis. Each subplot displays the % Contribution of MUA and LFP features for a specific rat ( Rat #4 , #6 , #9 , and #10 ) using the kSIR neural decoder. The contribution ratios of MUA features are shown individually for each channel (labeled 1–8), while the contribution ratios of LFP features are accumulated for each frequency band ( δ , θ , α , β , γ , and γ ′ ) across all channels. MUA features from each channel show the highest contribution in predicting both x- and y-velocity components compared to LFP power. Among the LFP frequency bands, the γ and γ ′ and exhibit higher contribution ratios, indicating their relevance in decoding forelimb movements. Data are presented as mean ± SD.

Article Snippet: Frequency-spectrum features were widely used for processing LFPs; therefore, LFP raw data were further down-sampled to a 1-kHz sampling rate and converted to power spectral density using a short-time Fourier transform with a Hanning window of 1 f m ms in length and time step of 33-ms, where f m is the minimum frequency of each LFP frequency bands (using the spectrogram function from the Signal Processing Toolbox, MATLAB R2019a, MathWorks).

Techniques: Labeling