A Novel Framework Combining CNN, LSTM, and Meta-Learning for Real-Time Monitoring of Anesthetic-Induced Brain States

• Published in: 2025 International Conference on Intelligent and Innovative Technologies in Computing, Electrical and Electronics (IITCEE) pp. 1 - 6
• Main Authors: G, Jyothi B, K, Arpitha
• Format: Conference Proceeding
• Language: English
• Published: IEEE 16.01.2025
• Subjects: Anesthesia; Anesthesia Monitoring; Brain modeling; Convolutional neural networks; Deep learning; Deep Learning in Anesthesia; Electroencephalogram (EEG); Electroencephalography; Levels of Consciousness; Long short term memory; Meta-Learning Framework; Metalearning; Monitoring; or signal to noise ratio; Real-time systems; Signal to noise ratio; SNR
• DOI: 10.1109/IITCEE64140.2025.10915368

In order to improve our understanding of brain function, monitoring the depth of unconsciousness during anesthesia is essential for both clinical practice and neuroscience research. The electroencephalogram (EEG) is a frequently used objective instrument for monitoring the changed states of the brain caused by anesthesia in real time. The effects of various general anesthetics on brain electrical activity are not the same. However, because of the low Signal to Noise Ratio (SNR) of EEG data, traditional machine learning models have difficulty with it, particularly in office-based anesthetic settings. Applications of Brain-Computer Interface (BCI) for pattern recognition and classification have seen a rise in the use of deep learning models, which are renowned for their great generalization and resilience in the face of noise. Deep learning has shown effective in a number of BCI applications, although its use in categorizing brain states while anesthesia is still relatively new. This research presents a unique framework to characterize brain states under anesthesia that combines deep neural networks and meta-learning. The framework uses a Long Short-Term Memory (LSTM) network to capture temporal relationships, Convolutional Neural Networks (CNN) to extract power spectrum features, and a meta-learning technique to address large cross-subject variability.