Deep anesthesia is associated with delayed recovery, postoperative cognitive dysfunction, and increased long-term mortality. In clinical practice, improper control of anesthesia depth can lead to intraoperative awareness or excessive suppression, thereby elevating the risk of postoperative neurological complications. Therefore, accurate monitoring and prediction of anesthesia depth are of great clinical importance. Currently, the bispectral index (BIS) is the most widely used clinical tool for assessing anesthesia depth by analyzing electroencephalogram (EEG) signals. However, existing BIS monitoring models can only provide real-time assessments and cannot effectively predict the trend of anesthesia depth. To address this limitation, this study proposes AnesNet, a deep learning–based model designed to provide early warnings of anesthesia depth 5, 10, and 15 min in advance. Considering diverse clinical needs, AnesNet supports both regression and classification prediction paradigms, enabling continuous depth forecasting and risk-oriented early warning of clinically unfavorable anesthesia states. A total of 2,803 surgical patients were included in this study. Results show that AnesNet achieved an AUROC of 0.895 in classification tasks and a MAE of 4.90 in regression tasks. Furthermore, feature ablation experiments demonstrated the model’s interpretability, enabling quantification of the contribution of individual physiological parameters to the prediction outcomes.
Qi et al. (Sat,) studied this question.