Abstracts

Rationale: Several methods assessing the complexity and entropy of biological signals can be used to investigate properties of the EEG during onset of partial seizures (PS). The relative efficiency of these methods is reported here. Methods: A subset of the most common methods assessing signal properties were selected for comparison. These include Shannon entropy (ShEn), spectral entropy (SpEn), sample entropy (SampEn), permutation entropy (PermH), Hjorth complexity, Lamberti complexity (CJS), LZC algorithmic complexity (Lempel-Ziv), signal complexity (GAD) and Higuchi fractal dimension (HFD). Each method was tested over a large set of partial seizures, both mesial temporal (MT), neocortical temporal (NeoT) and extra-temporal (NeoXT) onset, recorded from 45 consecutive patients with intracranial arrays. Normalized changes relative to an interictal baseline level were calculated for preictal periods and for ictal periods IC1 (0-8s) and IC2 (6-14s) after onset. Results: Changes observed during IC1 and IC2 periods are described here. Hjorth complexity shows a decrease for the MT group. LZC shows increases for all regions, the largest being for the NeoT group. GAD complexity exhibits early and fairly consistent increases for all regions. ShEn and SpEn are indicators of amplitude and frequency content and indicates the predominance of a frequency shift in the onset of PS from the NeoT group compared to seizures originating from other regions. HFD showed no changes for MT onset PS but marked increases for seizures originating from other regions. SampEn responses were very similar to LZC in all groups. PermH shows a marked decrease for MT onset PS but inconsistent results for other regions. CJS is not very specific for the early periods of the seizure IC1, and shows decreases later during IC2 except for the NeoT group. Measures can also be grouped by their relative responses. Figure 1 summarizes selected results for the MT and NeoT groups illustrating the relative ability of each method to detect the onset of PS and the ability to differentiate the region of onset. Each oval is centered on the mean level change occurring within 8 seconds of PS onset for MT onset patients (x-axis) and NeoT onset patients (y-axis). All measures are normalized to the standard deviation of a baseline period taken a minute prior to onset, and each oval horizontal and vertical dimension represent the standard error of the mean for the MT and NeoT group respectively. Methods whose oval is on the axis (x or y) cannot detect changes occurring at the onset of the seizures from MT or NeoT patients respectively. Oval on the diagonal cannot differentiate between MT and NeoT onset. Conclusions: The changes of properties of the EEG signals at the onset of PS are often complex and also involve changes in amplitude and frequency content. The most consistent measure to assess and therefore detect the onset of a partial seizure is the GAD signal complexity; however for the purpose of differentiation of the onset zone, a combination of GAD with other complexity (LZC, HFD) or entropy measures (SpEn, SampEn) might prove valuable. NIH NS48222