Title Information
Type of Resource (primo): dissertations
Name: Personal
Name: Personal
Name: Personal
Name: Personal
Name: Corporate
Origin Information
Physical Description
Note: thesis: Thesis (Sc. M.)--Brown University, 2022
Genre (aat): theses
Abstract: Objective: To develop a neural network for the binary classification of cardiac arrest patient electroencephalograms (EEGs) between two categories: general periodic discharges (GPDs) or non-GPDs. In addition, two different EEG montages, the longitudinal bipolar and the average reference, were used to train and test the model to see how montages affected classification. Methods: The dataset consisted of 14 neurologist-annotated EEGs from nine different cardiac arrest patients hospitalized at Rhode Island Hospital. The dataset contained 9 EEGs with GPDs and 5 EEGs with a diffusely slow background without GPDs. The data was filtered with a 60 Hz notch filter and a 70 Hz low pass filter before creating the montages and dividing the EEGs into 10 second snippets with a 50% overlap between each snippet. A 1-dimension convolutional neural network was designed and used for the task. One EEG with GPDs and one EEG without GPDs (non-GPDs) were selected to form the test set with the remaining 12 EEGs forming the training set; this process was repeated 45 times until all possible combinations of GPD and non-GPD pairs formed the test set. With 45 different train-test splits, 45 different models were created. Each model’s performance was evaluated using the following metrics: accuracy, sensitivity, specificity, precision, F1 score, false negative rate, and false positive rate. The final performance of the model was reported as the median of each respective metric from the models. This process was done for both montages. Results: Using the metrics described in the Methods section, the median accuracy, sensitivity, specificity, precision, and F1 score for the longitudinal bipolar montage were 97.46%, 100%, 98.13%, 97.39%, and 97.50% respectively. In the same order as the previous montage’s metrics were listed, the average reference montage’s median scores were 97.55%, 100%, 99.44%, 99.43%, and 97.62%. Conclusions: The comparable final metric scores of the two montages show the model’s potential at performing the task. However, much more EEG samples are needed for more generalization and to remove the imbalance between GPDs and non-GPDs. Future work includes exploring the frequency domain and testing other machine learning models (RNNs, 2D CNNs, etc.).
Subject
Subject (fast) (authorityURI="http://id.worldcat.org/fast", valueURI="http://id.worldcat.org/fast/00847042")
Language
Record Information