XAI-VSDoA: An Explainable AI-Based Scheme Using Vital Signs to Assess Depth of Anesthesia

Administration of anesthesia is essential in surgical procedures, ensuring patient unconsciousness and safety. Traditional Depth of Anesthesia (DoA) assessment methods rely heavily on the clinical expertise of anesthesiologists and patient physiological responses, which can vary widely due to age, weight, and ethnicity. This variability poses significant challenges in maintaining appropriate anesthesia levels and making timely decisions in critical situations. To address these challenges, we propose XAI-VSDoA, an explainable AI model using vital signs designed to augment DoA assessment by providing accurate predictions and interpretable insights. In this work, we experimented with various machine learning classifiers, including XGBoost, CatBoost, LightGBM, Random Forest, ResNet, and Feed-forward Neural Networks. Among these, the XGBoost model achieved the highest accuracy, with 99.34% on the University of Queensland dataset and 93.07% on the VitalDB dataset. Statistical testing confirmed that XGBoost outperformed the other models. We employed explainable AI techniques such as LIME and SHAP to identify the top 10 features significantly influencing the model’s predictions, ensuring the model’s transparency and reliability. These methods consistently highlighted the same influential features, reinforcing the model’s interpretability. Our proposed scheme demonstrated exceptional performance using numeric vital signs, with XAI techniques validating the key features. This interpretability boosts confidence in the model, enhancing its utility to augument and support the clininal observations of anethesiologiss in anesthesia management. Our findings underscore the potential of XAI-VSDoA as a valuable tool for clinical use, enhancing patient safety and decision-making in anesthesia.