Computational characteristics of interictal EEG as objective markers of epileptic spasms

Rachel J. Smith, Derek Hu, Daniel W. Shrey, Rajsekar R. Rajaraman, Shaun A. Hussain, Beth A. LopourbioRxiv 2020 · DOI

TL;DR

This study introduces a suite of computational EEG features that can reliably distinguish infants with epileptic spasms (ES) from healthy controls. By analyzing amplitude, spectral power, entropy, long-range temporal correlations, and functional connectivity, researchers developed a classifier that achieves 96% accuracy, paving the way for faster and more objective ES diagnosis.

What Problem Does This Paper Address?

Early diagnosis and prompt treatment are crucial for favorable neurodevelopmental outcomes in infants with epileptic spasms (ES). However, diagnosing ES can be challenging due to subtle clinical presentations, the wide variability of EEG patterns (like hypsarrhythmia), and low inter-rater reliability among clinicians. These diagnostic uncertainties can lead to delayed treatment and suboptimal outcomes. This paper aims to address this by identifying objective, computational EEG markers that can reliably differentiate ES patients from healthy controls, independent of the presence of hypsarrhythmia.

Methods

The researchers employed five computational electroencephalographic (EEG) measures:

  • Amplitude: Measured the overall voltage fluctuations in the EEG signal.
  • Power Spectra and Spectral Edge Frequency (SEF): Analyzed the distribution of power across different frequency bands (delta, theta, alpha, beta) and quantified the frequency below which 95% of the power resides.
  • Entropy (Shannon and Permutation): Quantified the complexity and predictability of the EEG signal. Shannon entropy focuses on the distribution of values, while permutation entropy considers the temporal order of the signal.
  • Long-Range Temporal Correlations (LRTC): Assessed using Detrended Fluctuation Analysis (DFA) to measure the strength of correlations over extended periods in the EEG signal. The DFA exponent ($\alpha$) and intercept ($\beta$) were calculated.
  • Functional Connectivity: Measured the correlation between EEG signals from different brain regions using both amplitude-based cross-correlation and phase-based phase-lag index (PLI).

These metrics were calculated from EEG data collected during both wakefulness and sleep. To mitigate bias, data were randomized and measurements were blinded. A logistic regression model was then used to combine the most discriminative features into a multivariable classifier.

Datasets

The study utilized EEG data from:

  • 40 ES patients: A diverse cohort, with the majority exhibiting refractory spasms.
  • 20 healthy control subjects: Initially, a group of children who underwent EEG to rule out ES but were found to be neurologically normal. An additional group of 25 seizure-free, neurologically normal children was added to create a more age-matched cohort.

The final cohort consisted of 40 ES cases and 20 control subjects, with an age-matched distribution.

Key Insights and Ideas

  • Amplitude as a General Biomarker: The finding of higher EEG amplitude in ES patients, even in the absence of hypsarrhythmia, suggests that amplitude itself can serve as a valuable, general biomarker for ES.
  • Entropy Reflects Brain Complexity: Lower entropy in ES patients indicates a less complex and potentially less organized brain state, which is a significant finding for understanding the underlying neurophysiology of the condition.
  • Functional Connectivity as a Key Differentiator: The consistently stronger functional connectivity networks observed in ES patients, particularly in the delta band, highlight disruptions in brain network communication as a core feature of the disorder.
  • Age-Dependent Effects: The study revealed that some EEG metrics become less discriminative in older children (over 4 years), suggesting a "pseudomaturation" effect where their EEG patterns start to resemble those of older normal controls. This is crucial for understanding the developmental trajectory of ES and its EEG markers.

Key Results

  • Amplitude: EEG amplitude was significantly higher in ES patients across all electrodes compared to controls.
  • Spectral Power: ES patients showed higher power in nearly all frequency bands, particularly in delta and theta frequencies during wakefulness.
  • Entropy: Shannon and permutation entropy were generally lower in ES patients compared to controls, indicating reduced signal complexity.
  • DFA Intercept: The DFA intercept ($\beta$) was significantly higher in ES patients across most frequency bands, suggesting greater variability in the amplitude envelope.
  • Functional Connectivity: ES patients exhibited significantly stronger functional connectivity networks than controls, both in wakefulness and sleep, across different frequency bands and connectivity measures.
  • Classification Accuracy: A multivariable logistic regression model combining sleep measures of functional connectivity (PLI), Shannon entropy, and DFA intercept achieved an Area Under the Curve (AUC) of 0.96, demonstrating high accuracy in distinguishing ES patients from controls.

Limitations and Future Work

  • Artifacts: Despite efforts to remove artifacts, the authors acknowledge that residual artifacts (e.g., from eye movements or muscle activity) might have influenced some results, particularly high frontal amplitude and temporal power.
  • Reference Choice: The choice of a linked-ear reference might have affected metric values, especially in central channels.
  • Sleep/Wake State Influence: The study noted that the discriminative power of some metrics varied between sleep and wakefulness. Future work could focus on identifying features consistently present across states or tailoring analysis based on the patient's state.
  • Age Dependency: The reduced classification accuracy in older children suggests that future algorithms might need to account for age-related changes or focus on younger populations.
  • Future Directions: The authors propose using these computational EEG markers to:
  • Identify pre-symptomatic patients.
  • Predict treatment response.
  • Measure treatment efficacy.
  • Predict relapse.
  • Investigate potential algorithms for early detection and intervention.

Summary updated

Abstract

Abstract Objective Favorable neurodevelopmental outcomes in epileptic spasms (ES) are tied to early diagnosis and prompt treatment, but uncertainty in the identification of the disease can delay this process. Therefore, we investigated five computational electroencephalographic (EEG) measures as markers of ES. Methods We measured 1) amplitude, 2) power spectra, 3) entropy, 4) long-range temporal correlations, via detrended fluctuation analysis (DFA) and 5) functional connectivity of EEG data from ES patients (n=40 patients) and healthy controls (n=20 subjects), with multiple blinded measurements during wakefulness and sleep for each patient. Results In ES patients, EEG amplitude was significantly higher in all electrodes. Shannon and permutation entropy were lower in ES patients than control subjects, while DFA intercept values in ES patients were significantly higher than control subjects. DFA exponent values were not significantly different between the groups. EEG functional connectivity networks in ES patients were significantly stronger than controls. Using logistic regression, a multi-attribute classifier was derived that accurately distinguished cases from controls (area under curve of 0.96). Conclusions Computational EEG features successfully distinguish ES patients from controls in a large, blinded study. Significance These objective EEG markers, in combination with other clinical factors, may speed the diagnosis and treatment of the disease, thereby improving long-term outcomes. Highlights Objective computational EEG features may aid diagnosis of epileptic spasms (ES) ES EEG has increased delta and theta power and decreased entropy relative to controls Stronger functional connectivity networks differentiate ES patients from controls
BibTeX
@misc{smith2020computational,
  author = {Rachel J. Smith and Derek Hu and Daniel W. Shrey and Rajsekar R. Rajaraman and Shaun A. Hussain and Beth A. Lopour},
  title = {{Computational characteristics of interictal EEG as objective markers of epileptic spasms}},
  year = {2020},
  doi = {10.1101/2020.11.13.380691},
  howpublished = {bioRxiv},
}
Figure