A Sedation Depth Assessment Model Based on Improved ResNet50-ECANet

ICU patients require sedation to manage their condition effectively. However, improper sedation levels can lead to fluctuations in circulatory function, worsen the patient's health status, and hinder the recovery process. Currently, sedation depth assessment in ICU patients primarily relies on subjective assessment by healthcare professionals and the use of brain function monitoring tools. This paper proposes an improved ResNet50-ECANet model for evaluating sedation depth by integrating the bispectral index (BIS) score from electroencephalography with objective analysis of physiological data. The process involves collecting ECG and PPG signals from patients, filtering out noise signals using wavelet techniques, and converting the processed signals into two-dimensional feature maps using the Gram angle field (GAF) encoding method. The ResN et50 architecture is utilized as the base network, with an improved ECANet incorporated to enhance the extraction of temporal features. Subsequently, the feature maps are fed into a Long Short Term Memory Network (LSTM) for temporal assessment of sedation depth, and the model's performance is validated using a custom dataset. Experimental results demonstrate an accuracy, recall, Fl score, and precision of 88.5%, 88.7%, 88.0%, and 0.878, respectively. The proposed algorithm effectively assesses sedation depth in ICU patients, meeting the assessment requirements of this patient population.