Preserving Motor Features by Alternative Re‐Referencing to Remove Heart Artifact on the Stentrode

Figures

• 

  FIGURE 1. Experimental design and sample vECoG recordings. Representation of the Stentrode, the motor mapping task and example recorded vECoG signals. (A) Diagram of the unfurled Stentrode, showing the placement of 16 electrodes along the scaffolding of the stent. (B) Diagram of Stentrode placement in the cerebral venous system, as well as the lead and the implantable receiver transmitter unit (IRTU). The green highlights the bilateral motor cortices. The blue represents the major cerebral veins, including the Superior Sagittal Sinus (rostral‐caudal) and the Superior Anastomotic Vein (medial‐lateral). The IRTU is seated in the chest of a participant and the lead connects it to the Stentrode. (C) Diagram of the motor mapping task the participants performed throughout this study. Participants are either seated or reclined during the task, and they are instructed which movement to execute before the run begins. (D) Sample recordings from channel 3 of P6 during the motor mapping task when right ankle movement is being performed with either the local or distal reference being used. (E) Log power spectral density of the channel illustrated in 1D, averaged over the 10 trials of rest and 10 trials…

• 

  FIGURE 2. Referencing scheme signal quality for vECoG data. (A) Both online hardware referencing schemes and offline software re‐referencing. Sample channels are taken from P6 during the motor mapping task with attempted right hand movement. Three examples of preserved ECG artifacts are highlighted in red on one channel for each referencing scheme. Recall that the local referencing scheme is a hardware reference, and thus the time series depicted is a different run than the one depicted for Distal, CAR and Laplacian referencing. Thus, the timing of the ECG artifacts in the Local referencing scheme is not expected to align with those in the other three schemes. (B) Power spectral density plots of the baseline recordings for P6 across sessions. The black dotted curve on each axis is the estimated aperiodic component of the average baseline PSD using FOOOF [45]. The solid, colored line on each axis represents the average baseline PSD with the specific referencing scheme, across sessions and channels. Motor cortical frequency bands of interest are highlighted along the x‐axis.

• 

  FIGURE 3. Band‐Limited Independent Component Analysis applied to vECoG data from P6. (A) The pipeline from raw vECoG data collected by the Stentrode to band‐limited independent components (BL‐ICs). The data shown is for bilateral ankle motor attempts in P6 with a distal monopolar reference which is subsequently band‐passed into alpha, beta, low gamma and high gamma bands. A band‐specific unmixing matrix is then constructed using FastICA [47], which reveals at least one component containing the electrocardiogram (ECG) artifact in the source space for alpha, beta and low gamma. (B) Example signals of distally referenced data (black) for bilateral ankle movements accompanied by several independent components resulting from ICA applied to the broadband signal. The task‐modulated signal is found on the independent component that captures the strongest ECG artifact, indicating broadband ICA may not be the best denoising technique. (C) The distribution of detected R‐R intervals on the isolated ECG components for all 5 participants. R‐R intervals were calculated using [48]. (D) The average detected ECG waveform for each participant, averaged across all trials and all ECG BLICs in alpha, beta and…

• 

  FIGURE 4. Task‐modulated independent components in the BL‐ICA source space. (A) The top task‐modulated BL‐IC for each cortical band of interest is shown as a time series during bilateral ankle movement in P6. (B) The average envelope of the top task‐modulated BL‐IC for the same run. (C) Difference in average power during rest and go trials for the run shown in (A, B). The four plots on the left visualize the power difference for the observed signals in alpha, beta, low gamma and high gamma for each channel. The fifth plot, on the right, shows the same calculation performed on the task‐modulated BL‐ICs. (D) beta burst detection on the top beta task‐modulated BL‐IC for bilateral ankle movement. Beta bursts are detected using the methodology explained in Section 2.8 in the source space. There is a clear decrease in beta burst frequency with the ‘Go’ cue, indicating that our task modulated beta BL‐IC not only preserves Event Related Desynchronization, but also the bursts relevant to bilateral ankle control.

• 

  FIGURE 5. Reconstruction and somatotopy. (A) Example time series of a channel in alpha (α), beta (β) and low gamma (𝛾low) bands before and after ECG removal with BL‐ICA. (B) Cross‐correlation analysis of channels before and after ECG artifact removal with BL‐ICA across all sessions. The normalized cross correlation of the channels was computed for the rest periods of each run in alpha, beta and low gamma, both before and after BL‐ICA. These correlation matrices were then averaged across all runs. (C) Channelwise relative entropy measurements in the alpha, beta and low gamma bands before and after cleaning with BL‐ICA. The dotted line represents relative entropy on each channel for the band‐passed, distally referenced signals. The colored points represent the same data after ECG components are removed through BL‐ICA. (D) The distribution of detected beta band burst amplitudes after BL‐ICA reconstruction.