Absence Seizure Control by a Brain Computer Interface

Abstract

The ultimate goal of epileptology is the complete abolishment of epileptic seizures. This might be achieved by a system that predicts seizure onset combined with a system that interferes with the process that leads to the onset of a seizure. Seizure prediction remains, as of yet, unresolved in absence-epilepsy, due to the sudden onset of seizures. We have developed a real-time absence seizure prediction algorithm, evaluated it and implemented it in an on-line, closed-loop brain stimulation system designed to prevent the spike-wave-discharges (SWDs), typical for absence epilepsy, in a genetic rat model. The algorithm corretly predicted 88% of the SWDs while the remaining were quickly detected. A high number of false-positive detections occurred mainly during light slow-wave-sleep. Inclusion of criteria to prevent false-positives greatly reduced the false alarm rate but decreased the sensitivity of the algoritm. Implementation of the latter version into a closed-loop brain-stimulation-system resulted in a 72% decrease in seizure activity. In contrast to long standing beliefs that SWDs are unpredictable, these results demonstrate that they can be predicted and that the development of closed-loop seizure prediction and prevention systems is a feasable step towards interventions to attain control and freedom from epileptic seizures.

Figure 1

(a) A set of EEG recordings taken from subgranular layers 4 (Ctx4) and 5 (Ctx5) of the somatosensory cortex and postero/lateral nucleus of the thalamus (PO). (b) Energy of wavelet transformation, corresponding to the EEG signals shown above and distributed over the range of timescales s = 1/f, where f is the linear frequency. (c) The resulting surface $W\left(s,t\right)={\prod }^{}{W}_i(s,t)$. The oscillatory pattern occurring prior to SWD onset, which is considered as a precursor of SWD, is encircled. (d) The momentary distributions of the wavelet energy in the 5–10 Hz band, taken for 4 seconds (I) and ~0.5 second (II) before SWD onset. The left figure shows that at 4 seconds before SWD onset only cortical EEGs exhibit a local synchronization in the precursor band (i.e. the local increase of energy appears in the frequency ~7 Hz). The wavelet energies of the cortical EEGs exhibit a synchronized increase (i.e. the lines are aligned to each other), while the wavelet energy of the thalamic EEG does not (i.e. the thalamic line is shifted relative to the others). The right figure shows that at ~0.5 second before SWD onset the wavelet energy is increased for all three channels. (e) Percentage of predicted and detected SWDs within the 4 hours recording of the first group of six WAG/Rij rats. Remember that no SWDs remained undetected by this (first) algorithm, the SWDs that were not predicted were quickly detected; Number of false positives across different states of alertness. For each rat and each state of alertness, 5 segments of 50 seconds duration were randomly selected in each recording for quantification of the number of false alarms.

Figure 2

(a) The experimental setup of the brain-computer interface. The set of analog inputs 1–6 of the acquisition hardware correspond to the three EEG channels (1-3), a marker for predictions (4), the stimulation pulse train of 1 sec (5) and the signal from the passive infrared registration system (PIR) for movement detection (6), respectively. The dashed line corresponds to the digital input of the PC, the feedback is shown by the shadow. (b) The prediction (upper case) and prevention (lower case) of the SWD/absence seizure by delivering a pulse train of 1.0 second duration. (c) Mean percentage of correctly predicted and correctly detected SWDs by the algorithm including two additional critera (see para on “On-line precursor detection”). The data are from an one hour baseline recording of 6 WAG/Rij rats of experiment 2; (d) Number of false positives for each individual rat generated by the revised algorithm within the 1 hour baseline of the 6 WAG/Rij rats (right panel), and the relative decrease in the false positives rate (left pannel) between the first algorithm (gray) and the second algorithm (red). Note, to determine the relative decrease, one hour EEG recordings of rats of the first six WAG/Rij rats were analyzed by the algorithm without the two additional criteria and the false positive counts of this analysis were taken as 100% reference. (e) The total duration of the epileptic activity and the behavioral activity during the baseline and the stimulation session, averaged over the group of rats. The error bars show the standart error of the mean of the group.