Complexity Modulation with Naturalistic Narrative Stimuli for Prognosis of Acute Brain-Injured Patients

Poster Session F - Tuesday, April 16, 2024, 8:00 – 10:00 am EDT, Sheraton Hall ABC

Hassan Al-Hayawi¹ (halhaya@uwo.ca), Geoffrey Laforge², Adrian Owen³; ¹Department of Psychology, Western University, London, ON, Canada, N6A 3K7, ²Brain and Mind Institute, Western University, London, ON, Canada, N6A 3K7, ³Department of Physiology and Pharmacology, Schulich School of Medicine and Dentistry, Western University, London, ON, Canada, N6A 3K7

The clinical utility of electroencephalography (EEG) to aid in the diagnosis and prognosis of brain-injured patients in the intensive-care unit (ICU) is promising. Specifically, advances have been made in machine learning and complexity EEG methods for classifying different cognitive states. This study aimed to evaluate the usefulness of EEG in predicting good neurologic recovery in unresponsive patients with severe brain injury in the ICU. The study recruited 33 ICU patients and divided them into good (n = 16) and poor outcome (n = 17) groups based on their Glasgow Outcome Scale-Extended scores at 3,6, and 12-month follow-ups after injury. Additionally, 18 healthy control participants were recruited. Various complexity and entropy measures were extracted from the signals as features. The features were analyzed statistically, and the success of features in classifying between intact and scrambled conditions was measured by various classifiers using a stratified 5-fold cross-validation technique. Healthy controls were used to find the measures and algorithms that can best discriminate between intact and scrambled. The top-performing measures and algorithms were determined using a Two-way repeated measures ANOVA. Results from healthy controls showed that Support Vector Classification and Linear Discriminant Analysis were the top-performing algorithms. Furthermore, Fractal Line Length index and Conditional Weighted Permutation Entropy were the top-performing complexity measures for discriminating between intact and scrambled. However, none of these models were able to predict patient prognosis, as the resulting accuracy scores were not correlated with patient outcomes. Further research is necessary to develop these techniques to accurately predict patient outcomes.

Topic Area: METHODS: Electrophysiology