A pilot on intelligence fusion for anesthesia depth prediction during surgery using frontal cortex neural oscillations

Anesthetic agents are widely used for their hypnotic and sedative effects as part of surgical procedures, but despite their widespread use there continues to be suboptimal dosing of the anesthetic to the patients, which necessitates more effective means of monitoring the depth of anesthesia (DoA) as anesthetic agents are administered. Effective means towards DoA monitoring could improve the optimal dosing for each patient to reduce the incidence awareness under general anesthesia and post-operative cognitive dysfunction; as well as reduce the incidence of complications associated with overdosing, such as hypertension. This work presents a novel pilot case study on an ongoing research around more effective means of DoA prediction, where patient-specific models are designed using a combination of signal processing and machine learning alongside electroencephalography (EEG) signals acquired from the frontal cortex. This particular case study investigates the use of various intelligence sources, i.e., machine intelligence, representing unsupervised feature extraction from a convolutional neural network (CNN), and expert-based intelligence via handcrafted features for the prediction of the DoA. It was seen that the handcrafted features provided the highest prediction accuracy across the various patient data, due to the ability to ‘bake-in’ prior knowledge regarding the physics of the process into the feature extraction process. The highest prediction accuracy was seen to be 86.5 ± 9.9 % for the LDA classification model upon pre-processing with the Linear Series Decomposition Learner (LSDL) algorithm. The fusion of both intelligence sources also provided an equivalent prediction accuracy similar to that of the hand-crafted features only.