Hybrid Deep Learning Model for Bispectral Index Estimation using EEG Signal

Abstract

Electroencephalography (EEG) plays a crucial role as a neurophysiological tool in anesthesia, offering invaluable insights into brain activity and responses during surgical procedures. Key to this role is the Bispectral Index (BIS), which serves to quantify consciousness and anesthesia depth. Maintaining optimal anesthesia levels is critical in surgery to improve patient comfort and mitigate risks such as intraoperative awareness and postoperative complications. However, traditional BIS monitors face limitations due to proprietary algorithms and high costs, restricting their global accessibility. To address these challenges, we present a novel deep learning framework that integrates convolutional neural networks (CNNs), bidirectional long short-term memory (BiLSTM) networks, and attention mechanisms. This framework harnesses the temporal information inherent in EEG signals to deliver continuous, real-time predictions of BIS, facilitating precise anesthesia depth monitoring throughout surgical procedures. Trained on a diverse dataset encompassing EEG signals, numerous anesthesia drugs and BIS our model demonstrates robust performance, surpassing current state-of-the-art methods in accuracy and reliability. Notably, its ability to adaptively consider operative time enhances its predictive capabilities in real-world surgical settings. Through extensive experimentation and analysis across diverse anesthesia scenarios, we validate the efficacy of our approach and advocate for its adoption in clinical practice. This research underscores the potential of our proposed framework to revolutionize anesthesia monitoring by providing a cost-effective and accessible alternative to traditional BIS monitors. By surpassing existing methodologies, our framework paves the way for broader implementation in healthcare, promising significant advancements in anesthesia management and patient care