What brain signals are suitable for feedback control of deep brain stimulation in Parkinson's disease?

Abstract

Feedback control of deep brain stimulation (DBS) in Parkinson's disease has great potential to improve efficacy, reduce side effects, and decrease the cost of treatment. In this, the timing and intensity of stimulation are titrated according to biomarkers that capture current clinical state. Stimulation may be at standard high frequency or intelligently patterned to directly modify specific pathological rhythms. The search for and validation of appropriate feedback signals are therefore crucial. Signals recorded from the DBS electrode currently appear to be the most promising source of feedback. In particular, beta-frequency band oscillations in the local field potential recorded at the stimulation target may capture variation in bradykinesia and rigidity across patients, but this remains to be confirmed within patients. Biomarkers that reliably reflect other impairments, such as tremor, also need to be established. Finally, whether brain signals are causally important needs to be established before stimulation can be specifically patterned rather than delivered at empirically defined high frequency.

Figure 1

Effect of deep brain stimulation (DBS) of subthalamic nucleus on the local field potential (LFP). (A) Power autospectrum of LFP recorded without stimulation. There is a large peak arrowed at 14 Hz. (B) Frequency–time log power spectrum of LFP. Power, as in (A), shown over the pass band of the amplifier (4–40 Hz). Red bars along the time axis denote periods of DBS at 2.0–3.0 V. Dyskinesias of the contralateral foot were noted at voltages of 2.0 V and above. Note suppression of spectral peak with stimulation ≥2.0 V, with evidence of a temporary increase in the power of the peak with stimulation at 1.5 V and a delayed return of the peak after stimulation at 3.0 V is terminated. (C) Timing and voltage of DBS. Adapted from Eusebio et al., with permission.

Figure 2

Bispectral analysis. Mean bispectrum from 13 subthalamic nuclei before (A) and after (B) levodopa administration. The central 2-D plot shows the mean bispectrum of the LFP signals as a function of frequencies f1 (x-axis, in Hz) and f2 (y-axis, in Hz). The level lines in the plot represent bispectrum values color coded as indicated in the color bar on the right (log transform of the average bispectrum, expressed in log arbitrary units, log AU). The mean power spectrum is also shown adjacent to each axis. The diagonal in the central plot defines the two regions of symmetry of the bispectrum. (A) Before levodopa administration, the arrows indicate the harmonic nonlinear correlation between the LFP rhythm in the low-beta range (13–20 Hz, dashed lines) and the LFP rhythm in the high-beta range (20–35 Hz, continuous line). This nonlinear correlation is evidenced by the bispectral peak (13–20 Hz, 13–20 Hz). Note that this bispectral peak appears broad due to the frequency variability between nuclei. Bispectral peaks are also present in other regions; (2–7 Hz, 2–7 Hz), (8–12 Hz, 8–12 Hz), and (2–7 Hz, 8–12 Hz), suggesting the presence of nonlinear correlations between different LFP rhythms in the off Parkinsonian state. (B) After levodopa administration, bispectral peaks are suppressed. The mean spectral peak in the high-beta range is therefore now independent of activity at lower frequencies. Adapted from Marceglia et al., with permission.

Figure 3

Closed-loop stimulation in a Parkinsonian patient with tremor. The bottom panel (F) shows the spectrogram of the LFP signal and demonstrates a low-frequency beta peak in yellow (original unfiltered LFP shown in D). The beta power in the form of beta root mean squared (RMS) is also displayed (E), along with the trigger threshold (horizontal line). Crossing of this threshold gave a positive trigger output that was sustained for a minimum of one second or until beta power dropped below threshold again. Panel (A) shows trigger output. Stimulation (1.5 mV, 100 μs, 130 Hz) was delivered while trigger output was positive during the first half of the recording, denoted by the red line (B). Trigger continued after this but did not result in stimulation. Accelerometer recording (C) demonstrates good tremor suppression during closed-loop mode (first half) with 26% reduction in stimulation triggering time. Note the increase in beta RMS during the second half of the recording when there is no stimulation. Previously unpublished data.