Computational characterization of the electroencephalogram in patients with infantile spasms syndrome

Author(s): Smith, Rachel June | Advisor(s): Lopour, Beth A. | Abstract: Infantile spasms (IS) is a potentially catastrophic epileptic encephalopathy that typically presents in children six to twelve months of age and is characterized by clusters of seizures consisting of abrupt muscle spasms. IS is associated with a host of comorbidities, high mortality rates, debilitating neurocognitive stagnation and psychomotor delay, and often progresses to other highly refractory forms of epilepsy. Diagnosis and clinical treatment decisions in IS are difficult due to the wide range of underlying etiologies and concomitant epilepsies. Patients with IS also exhibit a broad spectrum of electroencephalographic (EEG) morphologies, including a disorganized, high-amplitude pattern called hypsarrhythmia. Additionally, although the presence of hypsarrhythmia is often used as a diagnostic criterion for IS, there is low inter-rater reliability for identification of the pattern, and it is not a strong predictor of outcome. This lack of diagnostic accuracy and inefficient treatment response evaluation can delay successful treatment, which is associated with worse long-term outcomes. Computational EEG biomarkers of IS that are independent of the presence of hypsarrhythmia could supplement standard visual inspection of the EEG and enable objective identification of the disease and assessment of treatment response. We quantified basic characteristics of the pre-treatment EEG signal such as the amplitude, power spectrum, and Shannon entropy in cohorts of IS patients both with and without hypsarrhythmia. We identified significant differences between IS patients and age-matched control subjects, and these differences were robust to the presence of hypsarrhythmia. We also investigated the strength of long-range temporal correlations in the EEG with detrended fluctuation analysis (DFA) and developed statistical methods to infer confidence intervals for this metric. DFA differentiated IS patients from control subjects and reflected treatment response in post-treatment data. Lastly, we calculated EEG-based functional connectivity networks via cross-correlation and characterized long-term functional connectivity network changes in IS patients over multiple days. Analysis of normal control subjects allowed us to account for physiological fluctuations of the functional connections during sleep/wake cycles. In all, this work describes the pathological features of IS EEG data, providing an objective basis for diagnosis and laying the groundwork for early biomarkers of treatment response.