A Two-Stage Deep Learning Scheme to Estimate Depth of Anesthesia from EEG Signals

Abstract

Controlling the depth of anesthesia (DOA) through long surgeries is a crucial issue, and inaccurate dosage of pain killer and other anesthetic agents may lead to awareness or comma. Nonetheless, the accurate monitoring of DOA by analyzing electroencephalography (EEG) is still a challenge. To mimic the bispectral index (BIS) this study presents a deep learning method, which receives two EEG channels (located on the forehead) and continuously predicts the BIS score. The proposed method consists of convolutional neural network (residual network) followed by a recurrent neural network (bidirectional long short-term memory). In addition, we compare the performance of the proposed network with conventional methods in terms of regression and classification errors. All of the models are applied to a big dataset contains 176 subjects. The proposed network outperforms the conventional methods with respect to the generalization and both errors.