Correlation of EEG spectra, connectivity, and information theoretical biomarkers with psychological states in the epilepsy monitoring unit - A pilot study

Theta-Alpha Variability on Admission EEG Is Associated With Outcome in Pediatric Cerebral Malaria

PURPOSE

Pediatric cerebral malaria has high rates of mortality and neurologic morbidity. Although several biomarkers, including EEG, are associated with survival or morbidity, many are resource intensive or require skilled interpretation for clinical use. Automation of quantitative interpretation of EEG may be preferable in resource-limited settings, where trained interpreters are rare. As currently used quantitative EEG factors do not adequately describe the spectrum of variability seen in studies from children with cerebral malaria, the authors developed and validated a new quantitative EEG variable, theta-alpha variability (TAV).

METHODS

The authors developed TAV, a new quantitative variable, as a composite of multiple automated EEG outputs. EEG records from 194 children (6 months to 14 years old) with cerebral malaria were analyzed. Independent EEG interpreters performed standard quantitative and qualitative analyses, with the addition of the newly created variable. The associations of TAV with other quantitative EEG factors, a qualitative assessment of variability, and outcomes were assessed.

RESULTS

Theta-alpha variability was not highly correlated with alpha, theta, or delta power and was not associated with qualitative measures of variability. Children whose EEGs had higher values of TAV had a lower risk of death (odds ratio = 0.934, 95% confidence interval = 0.902-0.966) or neurologic sequelae (odds ratio = 0.960, 95% confidence interval = 0.932-0.990) compared with those with lower values. Receiver operating characteristic analysis in predicting death at a TAV threshold of 0.244 yielded a sensitivity of 74% and specificity of 70% for an area under the receiver operating characteristic curve of 0.755.

CONCLUSIONS

Theta-alpha variability is independently associated with outcome in pediatric cerebral malaria and can predict death with high sensitivity and specificity. Automated determination of this newly created EEG factor holds promise as a potential method to increase the clinical utility of EEG in resource-limited settings by allowing interventions to be targeted to those at higher risk of death or disability.

Normative Structure of Resting-State EEG in Bipolar Derivations for Daily Clinical Practice: A Pilot Study

We used numerical methods to define the normative structure of resting-state EEG (rsEEG) in a pilot study of 37 healthy subjects (10-74 years old), using a double-banana bipolar montage. Artifact-free 120-200 s epoch lengths were visually identified and divided into 1 s windows with a 10% overlap. Differential channels were grouped by frontal, parieto-occipital, and temporal lobes. For every channel, the power spectrum was calculated and used to compute the area for delta (0-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), and beta (13-30 Hz) bands and was log-transformed. Furthermore, Shannon's spectral entropy (SSE) and coherence by bands were computed. Finally, we also calculated the main frequency and amplitude of the posterior dominant rhythm. According to the age-dependent distribution of the bands, we divided the patients in the following three groups: younger than 20; between 21 and 50; and older than 51 years old. The distribution of bands and coherence was different for the three groups depending on the brain lobes. We described the normative equations for the three age groups and for every brain lobe. We showed the feasibility of a normative structure of rsEEG picked up with a double-banana montage.

EEG Emotion Recognition Based on Temporal and Spatial Features of Sensitive signals

Currently, there are some problems in the electrocorticogram (EEG) emotion recognition research, such as single feature, redundant signal, which make it impossible to achieve high-precision recognition accuracy when used a few channel signals. To solve the abovementioned problems, the authors proposed a method for emotion recognition based on long short-term memory (LSTM) neural network and convolutional neural network (CNN) combined with neurophysiological knowledge. First, the authors selected emotion-sensitive signals based on the physiological function of EEG regions and the active scenario of the band signals, and then merged temporal and spatial features extracted from sensitive signals by LSTM and CNN. Finally, merged features were classified to recognize emotion. The method was experimented on the DEAP dataset, the average accuracy in the valence and arousal dimensions were 92.87% and 93.23%, respectively. Compared with similar studies, it not only improved the recognition accuracy, but also greatly reduced the calculation channel, which proved the superiority of the method.