心肺繞道手術在臨床心臟手術中廣泛應用，例如冠狀動脈繞道手術每年執行次數超過六十萬例。然而，心肺繞道手術過程會引發全身性發炎、低血氧、再灌注傷害與微栓塞等生理變化，進而影響腦部功能，造成術後譫妄及長期認知功能變化。過往臨床多以紙筆測驗評估認知功能如簡易智能狀態測驗與蒙特婁認知評估，但這些方式受主觀性影響較大且需專業人員執行。相比之下，腦電圖能以高時間解析度提供客觀且非侵入性的腦功能資訊，並可藉由功率譜、複雜度及時域特徵計算作為潛在的神經標記。然而，腦電圖所能衍生的特徵數量龐大，若在資料集較小且未加以篩選特徵的情況下，容易導致模型維度過高、效率低落甚至過擬合，因此建立有效的特徵選擇方法顯得特別重要。本研究的目標為建立一個特徵選擇框架，將各種術前腦電圖特徵與術中因素結合，以預測術後認知變化並識別關鍵神經標記。

本研究蒐集44位接受心肺繞道手術患者之術前腦電圖、手術中臨床資料及術前術後蒙特婁測試分數，根據長期認知功能變化區分為術後上升、持平、下降三群，並透過特徵計算、統計分析與機器學習方法建立可以篩選重要特徵的模型。我們提出一個特徵選擇流程，結合不同類型腦電圖特徵與手術因子，進行特徵篩選與分類模型訓練。結果顯示，CatBoost模型在三分類的任務上達到61.36%的分類精準度，且複雜度與時域特徵（如近似熵、分形維度等）在beta與gamma頻帶多次被模型選為重要特徵，顯示其可作為預測術後認知變化的可靠神經標記。此外，消融實驗結果亦指出，結合特徵選擇流程的模型能有效提升分類準確度。

本研究證實，術前腦電圖特徵能有效預測術後認知功能變化且重要特徵或許有潛力應用於未來其他手術後的預測或疾病早期偵測。此外，本研究所提出的特徵選擇流程可從眾多特徵中篩選出關鍵特徵，進而提升模型在分類任務中的準確度，有利於其他針對小樣本資料集的分析，為其建構分類模型和篩選神經標記。

Cardiopulmonary bypass (CPB) is widely applied in clinical cardiac surgery, with over 600,000 coronary artery bypass grafting procedures performed annually. However, CPB induces systemic inflammation, hypoxemia, reperfusion injury, and microembolism, which may impair brain function and lead to postoperative delirium and long-term cognitive change. Cognitive function has traditionally been assessed using paper-and-pencil tests, such as the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA), but these methods are subjective and require trained personnel. In contrast, electroencephalography (EEG) provides objective and non-invasive information on brain function with high temporal resolution, and its power spectral, complexity, and time-domain features can serve as potential neuromarkers. However, the large number of features that can be derived from EEG presents challenges, particularly in small datasets, as unfiltered features can result in excessive dimensionality, low computational efficiency, and overfitting. Therefore, it is crucial to establish an effective feature selection strategy. The aim of this study was to develop a feature selection framework that integrates preoperative EEG features with intraoperative factors to predict postoperative cognitive change and identify key neuromarkers.

This study included 44 patients who underwent CPB surgery. Preoperative EEG, intraoperative factors, and MoCA scores obtained before and after surgery were collected. Patients were classified into three groups (improved, stationary, and declined) according to their long-term cognitive change. Feature computation, statistical analysis, and machine learning methods were applied to construct models capable of identifying important features. We proposed a feature selection framework that combines different types of EEG features with intraoperative factors for feature selection and classification. The results showed that the CatBoost model achieved a classification accuracy of 61.36% in the three-class classification. Moreover, complexity and time-domain features, such as approximate entropy and fractal dimension in the beta and gamma frequency bands, were repeatedly identified as important predictors by feature selection framework, suggesting their potential as reliable neuromarkers of postoperative cognitive change. In addition, ablation study demonstrated that the proposed feature selection framework effectively improved classification accuracy.

This study demonstrates that preoperative EEG features show potential to predict postoperative cognitive change and that important features may have potential applications in predicting change after other surgeries or in the early detection of neurological disorders. Furthermore, the proposed feature selection framework can identify key features from a large pool of candidates, thereby enhancing model accuracy in classification tasks. This approach is particularly beneficial for analyses based on small datasets, supporting both the construction of predictive models and the identification of neuromarkers.

[1] David P. Taggart. Coronary artery bypass surgery. Medicine, 50(7):445–448, July 2022.
[2] Yi Ming Zhuang, Ji Yang Xu, Kun Zheng, and Hong Zhang. Research progress of postoperative cognitive dysfunction in cardiac surgery under cardiopulmonary bypass. Ibrain, 10(3):290–304, August 2023.
[3] Marshal F. Folstein, Susan E. Folstein, and Paul R. McHugh. “Mini-mental state”: A practical method for grading the cognitive state of patients for the clinician. Journal of Psychiatric Research, 12(3):189–198, November 1975.
[4] Ziad S. Nasreddine, Natalie A. Phillips, Valérie Bédirian, Simon Charbonneau, Victor Whitehead, Isabelle Collin, Jeffrey L. Cummings, and Howard Chertkow. The Montreal Cognitive Assessment, MoCA: A Brief Screening Tool for Mild Cognitive Impairment. Journal of the American Geriatrics Society, 53(4):695–699, 2005.
[5] Khalil AlSharabi, Yasser Bin Salamah, Majid Aljalal, Akram M. Abdurraqeeb, and Fahd A. Alturki. EEG-based clinical decision support system for Alzheimer’s disorders diagnosis using EMD and deep learning techniques. Frontiers in Human Neuroscience, 17, August 2023.
[6] Irwin Feinberg, Richard L. Koresko, and Naomi Heller. EEG sleep patterns as a function of normal and pathological aging in man. Journal of Psychiatric Research, 5(2):107–144, June 1967.
[7] Xiaoli Yang, Zhipeng Fan, Zhenwei Li, and Jiayi Zhou. Resting-state EEG microstate features for Alzheimer’s disease classification. PLOS ONE, 19(12):e0311958, December 2024.
[8] Erol Başar and Bahar Güntekin. Review of delta, theta, alpha, beta, and gamma response oscillations in neuropsychiatric disorders. In Supplements to Clinical Neurophysiology, volume 62, pages 303–341. January 2013.
[9] Anruo Shen, Yunxia Li, Xiaorong Gao, Bai Lu, Jingnan Sun, and Yike Sun. An ensemble learning model for continuous cognition assessment based on resting-state EEG. npj Aging, January 2024.
[10] Akihiko Morita, Satoshi Kamei, and Tomohiko Mizutani. Relationship between slowing of the EEG and cognitive impairment in Parkinson disease. Journal of Clinical Neurophysiology, 28(4):384, August 2011.
[11] Mona Momeni, Sabrina Meyer, Marie-Agnès Docquier, Guillaume Lemaire, David Kahn, Céline Khalifa, Maria Rosal Martins, Michel Van Dyck, Luc-Marie Jacquet, André Peeters, and Christine Watremez. Predicting postoperative delirium and postoperative cognitive decline with combined intraoperative electroencephalogram monitoring and cerebral near-infrared spectroscopy in patients undergoing cardiac interventions. Journal of Clinical Monitoring and Computing, 33(6):999–1009, December 2019.
[12] Mariana Thedim, Duygu Aydin, Gerhard Schneider, Rajesh Kumar, Matthias Kreuzer, and Susana Vacas. Preoperative biomarkers associated with delayed neurocognitive recovery. Journal of Clinical Monitoring and Computing, 39(1):1–9, February 2025.
[13] Sérgio Daniel Rodrigues and Pedro Miguel Rodrigues. Electroencephalogram-based time-frequency analysis for Alzheimer’s disease detection using machine learning. Journal of Biological Methods, 12(1):e99010042, November 2024.
[14] Hao Jia, Zihao Huang, Cesar F. Caiafa, Feng Duan, Yu Zhang, Zhe Sun, and Jordi Solé Casals. Assessing the potential of data augmentation in EEG functional connectivity for early detection of Alzheimer’s disease. Cognitive Computation, 16(1):229–242, January 2024.
[15] Yongjie Xu, Zengjie Yu, Yisheng Li, Yuehan Liu, Ye Li, and Yishan Wang. Autism spectrum disorder diagnosis with EEG signals using time series maps of brain functional connectivity and a combined CNN–LSTM model. Computer Methods and Programs in Biomedicine, 250:108196, June 2024.
[16] Lin Bao, Taotao Liu, Zhenzhen Zhang, Qian Pan, Lifang Wang, Guohui Fan, Zhengqian Li, and Yiqing Yin. The prediction of postoperative delirium with the preoperative bispectral index in older aged patients: a cohort study. Aging Clinical and Experimental Research, 35(7):1531–1539, July 2023.
[17] Sean Tanabe, Rosaleena Mohanty, Heidi Lindroth, Cameron Casey, Tyler Ballweg, Zahra Farahbakhsh, Bryan Krause, Vivek Prabhakaran, Matthew I. Banks, and Robert D. Sanders. Cohort study into the neural correlates of postoperative delirium: the role of connectivity and slow-wave activity. BJA: British Journal of Anaesthesia, 125(1):55–66, July 2020.
[18] Leah Acker, Megan K. Wong, Mary C. Wright, Melody Reese, Charles M. Giattino, Kenneth C. Roberts, Sandra Au, Cathleen Colon-Emeric, Lewis A. Lipsitz, Michael J. Devinney, Jeffrey Browndyke, Sarada Eleswarpu, Eugene Moretti, Heather E. Whitson, Miles Berger, Marty G. Woldorff, and INTUIT and PRIME study groups. Preoperative electroencephalographic alpha-power changes with eyes opening are associated with postoperative attention impairment and inattention-related delirium severity. British Journal of Anaesthesia, 132(1):154–163, January 2024.
[19] Hairong Liu, Jing Wang, Xin Xin, Peng Wang, Wanting Jiang, and Tao Meng. The relationship and pathways between resting-state EEG, physical function, and cognitive function in older adults. BMC Geriatrics, 24:463, May 2024.
[20] E. Hauser, R. Seidl, D. Rohrbach, I. Hartl, M. Marx, and M. Wimmer. Quantitative EEG before and after open heart surgery in children. A significant decrease in the beta and alpha2 bands postoperatively. Electroencephalography and Clinical Neurophysiology, 87(5):284–290, November 1993.
[21] Shogo Mito, Miho Miyajima, Hirofumi Tomioka, Hitomi Sato, Takashi Takeuchi, Hitoshi Muto, Yuji Kabasawa, Hiroyuki Harada, Kana Eguchi, Shota Kato, and Manabu Kano. Postoperative delirium prediction based on preoperative electrocardiogram and electroencephalogram. In 2024 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC), pages 1–5, December 2024.
[22] Yu-Ning Hu, Tsung-Hao Hsieh, Meng-Ta Tsai, Chung-Yao Chien, Jun-Neng Roan, Yu-Ching Huang, and Sheng-Fu Liang. Cognitive function deterioration after cardiopulmonary bypass: can intraoperative optimal cerebral regional tissue oxygen saturation predict postoperative cognitive function? Journal of Cardiothoracic and Vascular Anesthesia, 37(5):715–723, May 2023.
[23] Sreemanti Dey and R. Michael Alvarez. Fuzzy forests for feature selection in high dimensional survey data: An application to the 2020 U.S. Presidential Election. International Conference on Applied Machine Learning and Data Analytics, March 2022.
[24] Ahmed Omar and Tarek Abd El-Hafeez. Optimizing epileptic seizure recognition performance with feature scaling and dropout layers. Neural Computing and Applications, 36(6):2835–2852, February 2024.
[25] Farzad Golnoori, Farsad Zamani Boroujeni, and Seyed Amirhassan Monadjemi. A comparative study on deep feature selection methods for skin lesion classification. IET Image Processing, 18(4):996–1013, 2024.
[26] G. H. Klem, H. O. Lüders, H. H. Jasper, and C. Elger. The ten-twenty electrode system of the International Federation. The International Federation of Clinical Neurophysiology. Electroencephalography and Clinical Neurophysiology. Supplement, 52:3–6, 1999.
[27] Rainer Wirth, Christiane Nicola Klimek, Gero Lueg, Maryam Pourhassan, Louisa Maria Danielzik, Caroline Krüger, and Ulrike Sonja Trampisch. Acute disease induced cognitive dysfunction in older patients – an unrecognized syndrome. BMC Geriatrics, 22(1):670, August 2022.
[28] Elias Lindvall, Tamar Abzhandadze, Terence J. Quinn, Katharina S. Sunnerhagen, and Erik Lundström. Is the difference real, is the difference relevant: the minimal detectable and clinically important changes in the Montreal Cognitive Assessment. Cerebral Circulation - Cognition and Behavior, 6:100222, May 2024.
[29] Panithi Pongraweewan, Surapa Tornsatitkul, Arunotai Siriussawakul, Vijay Krishnamoorthy, Nipaporn Sangarunakul, Chalita Jiraphorncharas, Solaphat Hemrungrojn, and Natawut Nupairoj. Incidence of postoperative cognitive dysfunction in older adults: a prospective cohort study using a web-based Montreal cognitive assessment application. Scientific Reports, 15:31180, August 2025.
[30] Rafał Nowicki, Mikołaj Berezowski, Julita Kulbacka, Katarzyna Bieżuńska-Kusiak, Marek Jasiński, and Jolanta Saczko. Custodiol HTK versus Plegisol: in-vitro comparison with the use of immature (H9C2) and mature (HCM) cardiomyocytes cultures. BMC Cardiovascular Disorders, 22(1):108, March 2022.
[31] Arnaud Delorme and Scott Makeig. EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis. Journal of Neuroscience Methods, 134(1):9–21, March 2004.
[32] Scott Makeig, Anthony J. Bell, Tzyy-Ping Jung, and Terrence J. Sejnowski. Independent component analysis of electroencephalographic data. Proceedings of the 9th International Conference on Neural Information Processing Systems, pages 145–151, November 1995.
[33] Ruofan Wang, Jiang Wang, Haitao Yu, Xile Wei, Chen Yang, and Bin Deng. Power spectral density and coherence analysis of Alzheimer’s EEG. Cognitive Neurodynamics, 9(3):291–304, June 2015.
[34] Yuta Iinuma, Sou Nobukawa, Kimiko Mizukami, Megumi Kawaguchi, Masato Higashima, Yuji Tanaka, Teruya Yamanishi, and Tetsuya Takahashi. Enhanced temporal complexity of EEG signals in older individuals with high cognitive functions. Frontiers in Neuroscience, 16, September 2022.
[35] Aimee A. Flores-Sandoval, Paula Davila-Pérez, Stephanie S. Buss, Kevin Donohoe, Margaret O’Connor, Mouhsin M. Shafi, Alvaro Pascual-Leone, Christopher S. Y. Benwell, and Peter J. Fried. Spectral power ratio as a measure of EEG changes in mild cognitive impairment due to Alzheimer’s disease: a case-control study. Neurobiology of Aging, 130:50–60, October 2023.
[36] Tianqi Chen and Carlos Guestrin. XGBoost: A scalable tree boosting system. Association for Computing Machinery, August 2016.
[37] Liudmila Prokhorenkova, Gleb Gusev, Aleksandr Vorobev, Anna Veronika Dorogush, and Andrey Gulin. CatBoost: unbiased boosting with categorical features. Association for Computing Machinery, January 2019.
[38] Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory. Neural Computation, 9(8):1735–1780, January 1997.
[39] Vernon J. Lawhern, Amelia J. Solon, Nicholas R. Waytowich, Stephen M. Gordon, Chou P. Hung, and Brent J. Lance. EEGNet: a compact convolutional neural network for EEG-based brain–computer interfaces. Journal of Neural Engineering, 15(5):056013, July 2018.
[40] Franco Scarselli, Marco Gori, Ah Chung Tsoi, Markus Hagenbuchner, and Gabriele Monfardini. The graph neural network model. IEEE Transactions on Neural Networks, 20(1):61–80, January 2009.
[41] Dominik Klepl, Fei He, Min Wu, Daniel J. Blackburn, and Ptolemaios Sarrigiannis. EEG-based graph neural network classification of Alzheimer’s disease: an empirical evaluation of functional connectivity methods. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 30:2651–2660, September 2022.
[42] Gabriel Rilling, Patrick Flandrin, and Paulo Gonçalves. On empirical mode decomposition and its algorithms. IEEE-EURASIP Workshop on Nonlinear Signal and Image Processing, June 2003.
[43] Akshansh Gupta, Dhirendra Kumar, Hanuman Verma, M. Tanveer, Andreu Perez Javier, Chin-Teng Lin, and Mukesh Prasad. Recognition of multi-cognitive tasks from EEG signals using EMD methods. Neural Computing and Applications, 35(31):22989–23006, November 2023.
[44] Shufang Li, Weidong Zhou, Qi Yuan, Shujuan Geng, and Dongmei Cai. Feature extraction and recognition of ictal EEG using EMD and SVM. Computers in Biology and Medicine, 43(7):807–816, August 2013.
[45] Ali H. Al-Nuaimi, Emmanuel Jammeh, Lingfen Sun, and Emmanuel Ifeachor. Complexity measures for quantifying changes in electroencephalogram in Alzheimer’s disease. Complexity, 2018(1):8915079, January 2018.
[46] Masahiro Hata, Yuki Miyazaki, Chie Nagata, Hirotada Masuda, Tamiki Wada, Shun Takahashi, Ryouhei Ishii, Shigeru Miyagawa, Manabu Ikeda, and Takayoshi Ueno. Predicting postoperative delirium after cardiovascular surgeries from preoperative portable electroencephalography oscillations. Frontiers in Psychiatry, 14, November 2023.