Forecasting seizures in dogs with naturally occurring epilepsy

Abstract

Seizure forecasting has the potential to create new therapeutic strategies for epilepsy, such as providing patient warnings and delivering preemptive therapy. Progress on seizure forecasting, however, has been hindered by lack of sufficient data to rigorously evaluate the hypothesis that seizures are preceded by physiological changes, and are not simply random events. We investigated seizure forecasting in three dogs with naturally occurring focal epilepsy implanted with a device recording continuous intracranial EEG (iEEG). The iEEG spectral power in six frequency bands: delta (0.1-4 Hz), theta (4-8 Hz), alpha (8-12 Hz), beta (12-30 Hz), low-gamma (30-70 Hz), and high-gamma (70-180 Hz), were used as features. Logistic regression classifiers were trained to discriminate labeled pre-ictal and inter-ictal data segments using combinations of the band spectral power features. Performance was assessed on separate test data sets via 10-fold cross-validation. A total of 125 spontaneous seizures were detected in continuous iEEG recordings spanning 6.5 to 15 months from 3 dogs. When considering all seizures, the seizure forecasting algorithm performed significantly better than a Poisson-model chance predictor constrained to have the same time in warning for all 3 dogs over a range of total warning times. Seizure clusters were observed in all 3 dogs, and when the effect of seizure clusters was decreased by considering the subset of seizures separated by at least 4 hours, the forecasting performance remained better than chance for a subset of algorithm parameters. These results demonstrate that seizures in canine epilepsy are not randomly occurring events, and highlight the feasibility of long-term seizure forecasting using iEEG monitoring.

Figure 1. Seizure Advisory System (SAS) in Canines with Epilepsy.

(A) The implantable device for recording and storing continuous iEEG includes: Implantable Lead Assembly (ILA) placed in the subdural space (right), Implantable Telemetry Unit (ITU), and Personal Advisory Device (PAD). The system acquires 16 channels of iEEG and wirelessly transmits the data to the PAD. Data is stored on a flash drive and uploaded weekly via internet to a central data storage site. (B) Sixteen channels of intracranial EEG recorded with SAS. A focal onset, secondarily generalized seizure is shown. The top 1–8 channels are from the left hemisphere and 9–16 from the right hemisphere, as shown on the brain schematic above. The onset of the seizure is from left hemisphere (underlined) electrodes 3 & 4. (C) Schematic of temporal profile of forecast seizure probability and threshold defining the pre-ictal state, i.e. period of increased seizure probability. When the forecast probability exceeds the defined threshold a fixed duration warning is triggered. Consecutive warnings that occur within the duration of a prior warning are combined, allowing for variable duration warnings. i) Single warning triggered without seizure (false positive warning). ii) Multiple consecutive warnings combined into prolonged warning without seizure (false positive warning). iii) Compounded warning prior to seizure onset (true positive seizure warning prior to electrographic seizure onset).

Figure 2. Temporal profile of seizures.

(A) Full temporal record for canine 002, showing time of occurrence of the 27 clinically verified seizures (vertical red lines). The seizures fall into 5 clusters; each cluster is annotated with the number of seizures in the cluster. (B) Temporal profile of 96 power-in-band (PIB) features for canine 002, spanning approximately 2.8 days in the vicinity of seizure cluster 2 in (A). Each grouping of 6 traces shows the 6 PIB features derived from one iEEG recording channel; from bottom to top these features capture the power-in-band of the delta (δ: 0.1–4 Hz), theta (θ: 4–8 Hz), alpha (α: 8–12 Hz), beta (β: 12–30 Hz), gamma-low (γ-low: 30–70 Hz) and gamma-high (γ–high: 70–180 Hz) frequency bands, respectively. The grouping for channel 1 is at the bottom of the figure, and channel 16 at the top. Vertical red lines locate the occurrence of individual seizures. Light red shading indicates the 90-minute period preceding each seizure, which was labeled as pre-ictal for training purposes; everything else (white background) was treated as the inter-ictal state. (C) Expansion of traces for channels 9–12 in (B.

Figure 3. Seizure probability versus time.

(A) The seizure probability for 1-minute blocks of full iEEG record of canine 002, as predicted by a logistic regression classifier trained on 96 power-in-band (PIB) features. Green and red dots indicate blocks labeled inter-ictal and pre-ictal, respectively, for classifier training, and vertical blue lines indicate occurrences of clinically verified seizures. Gaps in the plot correspond to periods of invalid data. The staircase underneath the plot delineates the 10 cross-validation partitions used during classifier training and testing. (B) Seizure probabilities for 1-minute blocks of iEEG record of canine 002 in vicinity of seizure cluster 2 (second from left in Figure 3 (A)), as predicted by a logistic regression classifier trained on 96 power-in-band (PIB) features. Green and red dots indicate blocks labeled inter-ictal and pre-ictal, respectively, for classifier training, and vertical blue lines indicate occurrences of clinically verified seizures.