Predicting Recovery from Coma Following Cardiac Arrest with a Reduced Set of EEG Channels

The aim of our work (Univ_Pittsburgh) was to explore the feasibility of using a convolutional neural network (CNN), with a reduced set of EEG channels, fused with a Random Forest to predict coma patient outcomes. This work is part of the 'Predicting Neurological Recovery from Coma After Cardiac...

Full description

Saved in:
Bibliographic Details
Main Authors: Riek, Nathan T, Elmer, Jonathan, Al-Zaiti, Salah, Akcakaya, Murat
Format: Conference Proceeding
Language:English
Subjects:
Brain modeling
Cardiac arrest
Computational modeling
Computer architecture
Electrodes
Electroencephalography
Training
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The aim of our work (Univ_Pittsburgh) was to explore the feasibility of using a convolutional neural network (CNN), with a reduced set of EEG channels, fused with a Random Forest to predict coma patient outcomes. This work is part of the 'Predicting Neurological Recovery from Coma After Cardiac Arrest: The George B. Moody PhysioNet Challenge 2023'. A 1D-CNN based on the ResNet-18 model was used to detect specific patterns in the EEG unique to either a good or poor outcome for the coma patient. To reduce dimensionality, electrodes were grouped into 5 regions. The CNN was fused with a Random Forest trained on patient features. The CNN and Random Forest model achieved True Positive Rates (TPRs) of 0.50+/-0.09 using 5-fold cross validation within the training set, 0.809 on the training set, 0.448 on the validation set, and 0.530 on the test set. Our team ranked 14th out of 36 teams. This work demonstrated the feasibility of grouping EEG channels to reduce dimensionality in the prediction of coma recovery. Future work should explore the use of different model architectures with the reduced set of EEG channels to achieve even higher performance.
ISSN:2325-887X
DOI:10.22489/CinC.2023.044