TL;DR
This study demonstrates that analyzing high-frequency oscillations (HFOs) in scalp EEG recordings can serve as an objective biomarker for infantile spasms, outperforming the traditional hypsarrhythmia measure. The findings suggest that increased HFO rates and their coupling with slow-wave activity are strongly associated with the presence of active spasms, paving the way for more accurate and timely diagnoses.
What Problem Does This Paper Address?
Infantile spasms (IS) are a severe epileptic syndrome in infants that, if not treated promptly, can lead to significant neurodevelopmental deficits. Early diagnosis is crucial, and scalp EEG is a primary diagnostic tool. However, the current standard for diagnosing IS relies on identifying "hypsarrhythmia," a pattern in interictal EEG that is characterized by high voltage, disorganization, and multifocal epileptiform discharges. The problem is that the definition of hypsarrhythmia is subjective and suffers from poor inter-rater reliability, even among experienced clinicians. This subjectivity can lead to diagnostic delays and challenges. Therefore, there's a pressing need for more objective and accurate EEG biomarkers for infantile spasms.
Methods
The researchers employed a retrospective cohort study design, analyzing overnight slow-wave sleep EEGs from 30 infants suspected of having infantile spasms. The key methodological innovation lies in their approach to analyzing high-frequency oscillations (HFOs), which are defined as oscillatory events between 80-500 Hz. They measured two main HFO-related metrics:
- HFO Rate: The frequency of HFO events per minute. This was initially identified using an automated short-term energy detector and then visually confirmed to filter out false positives like artifacts.
- Cross-Frequency Coupling (CFC): The strength of the relationship between the amplitude of HFOs and the phase of slow-wave activity (SWA) at specific frequencies (3-4 Hz and 0.5-1 Hz). This was quantified using modulation indices (MIs) computed via the EEGLab PACT plugin.
They then used logistic regression models to assess the predictive power of these HFO-related metrics and hypsarrhythmia for the presence of active epileptic spasms.
Datasets
The study utilized retrospective overnight video-EEG recordings from 30 infants suspected of having infantile spasms, collected at UCLA between October 2016 and May 2019. These recordings were sampled at 2,000 Hz. A subset of 8 children with serial EEG recordings was used for validation.
Key Insights and Ideas
- HFOs as a Reliable Biomarker: The study strongly suggests that scalp-recorded HFOs are a more objective and reliable biomarker for infantile spasms than the traditionally used hypsarrhythmia. This is a significant step towards standardizing diagnosis.
- Automated Analysis Potential: While visual verification of HFOs was performed, the study found that automated HFO detection alone achieved comparable predictive performance to models that included visual verification, highlighting the potential for efficient and reproducible analysis.
- Treatment Response Reflected in HFOs: A significant decrease in HFO rate and MI values was observed in infants who responded to treatment, indicating that HFOs might also serve as a marker for treatment efficacy.
- Fast Ripples on Scalp EEG: This study is among the first to report the detection of fast ripples (250-500 Hz) on scalp EEG in infants with infantile spasms, expanding the range of HFOs that can be utilized for diagnosis.
Key Results
- High Prevalence of Spasms: Out of 30 infants, 23 (77%) exhibited active epileptic spasms during the overnight EEG.
- HFOs Differentiate Spasm Groups: Increased HFO rates and MI values were significantly higher in children with active epileptic spasms compared to those without (p < 0.001 for HFO rate, p < 0.01 for MIs at 3-4 Hz, and p = 0.02 for MIs at 0.5-1 Hz).
- Superior Predictive Power: Logistic regression models incorporating HFO-related metrics achieved a significantly higher Area Under the Curve (AUC) (0.80–0.98) for predicting active spasms compared to models using hypsarrhythmia (AUC: 0.61).
- High Accuracy: The best predictive model, using HFO rate with visual verification, achieved an accuracy of 93% and was validated with 87.5% accuracy on a separate dataset.
- Automated Analysis Efficiency: Automated HFO detection alone provided strong predictive performance (AUC 0.96) with a significantly reduced analysis time (6.7 minutes per case) compared to visual verification (32.9 minutes per case).
Limitations and Future Work
- Small Sample Size: The study included a relatively small number of subjects (n=30), which the authors acknowledge limits the creation of more complex multivariable regression models and may affect external validity. Future work should aim to validate findings in larger cohorts.
- Sleep EEG Focus: The analysis was primarily conducted on sleep EEGs, which are less prone to artifacts. The authors suggest that developing methods to accurately analyze awake EEGs, which are easier to obtain, is a crucial next step.
- Artifact Correction: While visual verification helped filter artifacts, further research is needed to develop robust automated methods for artifact correction in awake EEGs.
- False-Negative Rate: The study did not quantify the false-negative rate of the HFO detector, meaning some true HFOs might have been missed.
- Localization Ambiguity: The study found that scalp EEG HFOs did not always clearly lateralize to the epileptogenic lesion, suggesting that HFOs might sometimes "falsely lateralize," similar to epileptic spikes. Further research is needed to clarify the localization value of scalp EEG HFOs.
- Physiological HFOs: The nature of physiological HFOs detected in neurologically normal children warrants further investigation.
Summary updated