@article{GONG2023, 
author = {Yusheng GONG and Wenhua WANG and Min PAN and Rui ZHANG},
title = {A novel fusion feature extraction method for prolonged disorders of consciousness assessment},
year = {2023},
journal = {Journal of Northwest University (Natural Science Edition)},
volume = {53},
number = {3},
pages = {367-376},
keywords = {prolonged disorders of consciousness (PDOC), electroencephalogram (EEG), sleep stage, fusion feature, back propagation neuron netwok},
url = {https://www.sciopen.com/article/10.16152/j.cnki.xdxbzr.2023-03-006},
doi = {10.16152/j.cnki.xdxbzr.2023-03-006},
abstract = {Prolonged disorders of consciousness (PDOC) is a kind of neurological disease caused by severe brain damage, which results in loss of consciousness more than 28 days. Vegetative state (VS), minimally conscious stateplus (MCS+) and minimally conscious state minus (MCS-) are the main states of prolonged disorders of consciousness. PDOC assessment is helpful for formulating reasonable treatment plans, promoting recovery of consciousness in PDOC patients. In clinic, the main assessment methods are behavioral rating scales, imaging examinations and so on. But above methods are impossible for continuously monitoring patients’ consciousness in real-time. Electroencephalogram (EEG) can record the state of brain’s electrical activity reflecting the level of consciousness in real-time. Moreover, sleep EEG is closely related to the level of consciousness in PDOC patients. Therefore, a novelfusion feature extraction method for PDOC assessment is proposed based on EEG data. Firstly, the PDOC-EEG signals are preprocessed by data expanding, filtering and denoising. Secondly, the pathological features of PDOC-EEG are extracted using time-domain, frequency-domain and nonlinear analysis. In this process, two new features, power weight-based channel coherence and dispersion of frequency correlation are designed. Meanwhile, consider the proportion of sleep stage duration as feature, which is related to the level of consciousnes, using EEG, EOG (electro-oculogram) and EMG (electromyogram) data. Then, fuse the extracted features. Finally, the automatic PDOC assessment is realized using random forest and back propagation neural network classifier. This method is verified by dataset of PDOC patients collected in the federal research and clinical center of intensive care medicine and rehabilitology in Russia. Numerical experiment shows that accuracy and sensitivity of the proposed method are 94.2% and 94.1%.}
}