Coma Outcome Predictive Analysis: Identification of suitable indicators with a main focus on brain connectivity

This thesis investigates the potential of EEG indicators, with a main focus on brain connectivity indicators, in predicting coma outcomes, particularly in patients with traumatic brain injury (TBI). EEG data from thirteen patients were analyzed, focusing on the extraction of features related to brain connectivity and their classification using various machine learning models. The methodology involved three approaches: a non-standard-function-based method and two standard-function-based methods, tailored and one-size-fits-all. Key indicators such as Partial Directed Coherence, Direct Transfer Function, and Granger Causality were extracted and analyzed. Results showed that the k-Nearest Neighbors (k-NN) classifier yielded the best performance, particularly when using the tailored standard-function-based approach. The study highlights the importance of connectivity measures in accurately predicting patient outcomes, although it was limited by the small sample size and the use of only four EEG channels. It is suggested that expanding the dataset and using more EEG channels could improve the robustness of predictive models, providing a foundation for further research in improving coma outcome prediction.