The Perioperative Neurocognitive Disorder Prediction Based on AI-Assisted EEG Dynamic Features in Anesthetized Mice

Background: Postoperative neurocognitive disorders (PND) are frequent complications in the elderly surgical patients, with aging recognized as a major risk factor. This study aimed to identify electrophysiological markers and establish an exploratory machine learning framework for PND-related vulnerability prediction using anesthetic electroencephalography (EEG) features in aged mice. Methods: Young and aged mice underwent laparotomy under isoflurane anesthesia with EEG recording. Neurocognitive performance was quantified by 16 standardized behavioral fractions. A semi-supervised K-means algorithm, anchored on young-surgery mice, stratified aged-surgery mice into PND and non-PND clusters. EEG dynamics during anesthesia maintenance and emergence were analyzed, and machine learning models were trained to predict PND from EEG features. Results: At baseline, neurocognitive function was comparable across groups. After anesthesia/surgery, aged mice exhibited selective spatial and contextual memory impairments, with two-thirds classified as PND. During emergence, PND mice displayed elevated δ power and reduced α and β ratios. A Multi-layer Perceptron classifier showed discriminatory performance for PND classification in one evaluation setting (AUC = 0.94). Conclusions: This study identifies emergence-related EEG features associated with postoperative neurocognitive vulnerability in aged mice and provides an exploratory machine learning framework for preclinical risk stratification. These findings support further mechanistic investigation and warrant future validation in human perioperative EEG datasets.