在這項研究中，我們提出基於EEG的卷積長短期記憶(Convolutional Long Short-Term Memory, ConvLSTM)網路，該網路利用深度學習技術對EEG訊號提取特徵來區分注意力不足過動症(Attention Deficit Hyperactivity Disorder, ADHD)和正常對照組(Neurotypical, NT)兒童。ADHD是一種神經發展障礙，從兒童時期就會有症狀的影響。近年許多研究為了解決臨床上的主觀性，使用腦電圖(Electroencephalography, EEG)的事件相關電位(Event-Related Potential, ERP)來辨識ADHD和NT之間的差異。我們則使用一種名為梯度加權類激活映射(Gradient-weighted Class Activation Mapping, Grad-CAM)的可視化技術來檢測腦電圖的特異性。共有36名ADHD兒童和24名NT兒童參加幼兒版持續性表現測驗(Kiddie-Continuous Performance Test, K-CPT)測試，同時接受腦電波監測。實驗結果表明，與卷積神經網路和長短期記憶網路相比，基於EEG的ConvLSTM網路產生了最佳性能，其準確率為86.16%。最後，將Grad-CAM觀察到的O2的delta功率和FP1、FP2和Fz的beta功率視為重要特徵，並使用支持向量機 (Support Vector Machine, SVM) 獲得平均80%準確率，該結果驗證深度學習特徵提取的能力取代手動的可行性。此外我們還比較了過去文獻所提出的特徵集之間的性能差異。這些發現表明提出的基於EEG的ConvLSTM網路模型具有深度學習技術提取特徵的能力，可以客觀的幫助ADHD臨床上的診斷。

In this study, an EEG-based convolutional long-short term memory (ConvLSTM) networks that utilized deep learning techniques is proposed to extract features from electroencephalography (EEG) signals to distinguish between Attention Deficit Hyperactivity Disorder (ADHD) and Neurotypical (NT) children. ADHD is a neurodevelopmental disorder that affects children with symptoms from childhood. In recent years, in order to address clinical subjectivity, many studies have used the Event-Related Potential (ERP) of EEG to identify differences between ADHD and NT groups. A visualization technique termed gradient-weighted class activation mapping (Grad-CAM) was employed to detect EEG specificities. A total of 36 children with ADHD and 24 neurotypical children participated in the Kiddie-Continuous Performance Test (K-CPT) while monitoring EEG. The experimental results showed that the EEG-based ConvLSTM networks produced the best performance with the accuracy of 86.16% in comparison with the convolutional neural networks and long-short term memory networks. Finally, the Support Vector Machine (SVM) model obtains an average accuracy of 80%; which considering delta power in O2 region and beta power in FP1, FP2 and Fz region as significant features obtained from the observation of Grad-CAM. This result validated the feasibility of deep learning to replace manual feature extraction. In addition, the performance differences between feature sets proposed in the past literatures are also compared. These findings show that the proposed EEG-based ConvLSTM networks with the ability of deep learning techniques to extract features can objectively help in the diagnosis of ADHD clinically.

Reference
[1] World Health Organization, "The ICD-10 classification of mental and behavioural disorders: clinical descriptions and diagnostic guidelines," World Health Organization, 1992.
[2] A.P. Association, "Diagnostic and statistical manual of mental disorders(DSM-5)," American Psychiatric Association, 2013.
[3] S.M. Snyder, T.A. Rugino, M. Hornig, and M.A. Stein, "Integration of an EEG biomarker with a clinician's ADHD evaluation," Brain and Behavior, vol. 5, no. 4, 2015.
[4] W. Samek, A. Binder, G. Montavon, S. Lapuschkin, and K.R. Müller, "Evaluating the visualization of what a deep neural network has learned," IEEE Transactions on Neural Networks and Learning Systems, vol. 28, no. 11, pp. 2660-2673, 2016.
[5] H.H. Jasper, P. Solomon, and C. Bradley, "Electroencephalographic analyses of behavior problem children," The American Journal of Psychiatry, vol. 95, no. 3, pp. 641-658, 1938.
[6] S.K. Loo, A. Cho, T.S. Hale, J. McGough, J. McCracken, and S.L. Smalley, "Characterization of the theta to beta ratio in ADHD: identifying potential sources of heterogeneity," Journal of Attention Disorders, vol. 17, no. 5, pp. 384-392, 2012.
[7] T. Shi, X. Li, J. Song, N. Zhao, C. Sun, W. Xia, L. Wu, and A. Tomoda, "EEG characteristics and visual cognitive function of children with attention deficit hyperactivity disorder (ADHD)," Brain & Development, vol. 34, no. 10, pp. 806-811, 2012.
[8] M. Arns, C.K. Conners, and H.C. Kraemer, "A decade of EEG theta/beta ratio research in ADHD: a meta-analysis," Journal of Attention Disorders, vol. 17, no. 5, pp. 374-383, 2013.
[9] M. Bink, G.J.M. van Boxtel, A. Popma, I.L. Bongers, A.J.M. Denissen, and C. van Nieuwenhuizen, "EEG theta and beta power spectra in adolescents with ADHD versus adolescents with ASD + ADHD," European Child & Adolescent Psychiatry, vol. 24, no. 8, pp. 873-886, 2014.
[10] S.M. Snyder, H. Quintana, S.B. Sexson, P. Knott, A.F.M. Haque, and D.A. Reynolds, "Blinded, multi-center validation of EEG and rating scales in identifying ADHD within a clinical sample," Psychiatry Research, vol.159, no. 3, pp. 346-358, 2008.
[11] A.R. Clarke, R.J. Barry, F.E. Dupuy, L.D. Heckel, R. McCarthy, M. Selikowitz, and S.J. Johnstone, "Behavioural differences between EEG-defined subgroups of children with attention-deficit/hyperactivity disorder," Clinical Neurophysiology, vol. 122, no. 7, pp. 1333-1341, 2011.
[12] A. Tenev, S. Markovska-Simoska, L. Kocarev, J. Pop-Jordanov, A. Müller, and G. Candrian, "Machine learning approach for classification of ADHD adults," International Journal of Psychophysiology, vol. 93, no. 1, pp. 162-166, 2014.
[13] J. Anuradha, Tisha, V. Ramachandran, K.V. Arulalan, and B.K. Tripathy, "Diagnosis of ADHD using SVM algorithm," Proceedings of the Third Annual ACM Bangalore Conference, no. 29, pp. 1-4, 2010.
[14] H.T. Tor, C.P. Ooi, N.SJ. Lim-Ashworth, J.K.E. Wei, V. Jahmunah, S.L. Oh, U.R. Acharya, and D.S.S. Fung, "Automated detection of conduct disorder and attention deficit hyperactivity disorder using decomposition and nonlinear techniques with EEG signals," Computer Methods and Programs in Biomedicine, vol. 200, 2021.
[15] M.R. Mohammadi, A. Khaleghi, A.M. Nasrabadi, S. Rafieivand, M. Begol, and H. Zarafshan, "EEG classification of ADHD and normal children using non-linear features and neural network," Biomedical Engineering Letters, vol. 6, no. 2, pp. 66–73, 2016.
[16] Y. Chang, C. Stevenson, I.C. Chen, D.S. Lin, and L.W. Ko, "Neurological state changes indicative of ADHD in children learned via EEG-based LSTM networks," Journal of Neural Engineering, vol. 19, no. 1, 2022.
[17] L. Dubreuil-Vall, G. Ruffini, and J.A. Camprodon, "Deep Learning Convolutional Neural Networks Discriminate Adult ADHD From Healthy Individuals on the Basis of Event-Related Spectral EEG," Frontiers in Neuroscience, vol. 14, 2020.
[18] D.E. Rumelhart, G.E. Hinton, and R.J. Williams, "Learning representations by backpropagating errors," Nature, vol. 323, pp. 533-536, 1986.
[19] G. Ruffini, D. Ibañez, M. Castellano, S. Dunne, and A. Soria-Frisch, "EEG-driven RNN classification for prognosis of neurodegeneration in at-risk patients," 2016 25th International Conference on Artificial Neural Networks, pp. 306-313, 2016.
[20] S. Hochreiter, and J. Schmidhuber, "Long short-term memory," Neural Computation, vol. 9, no. 8, pp. 1735-1780, 1997.
[21] S. Alhagry, A.A. Fahmy, and R.A. El-Khoribi, "Emotion recognition based on EEG using LSTM recurrent neural network," International Journal of Advanced Computer Science and Applications, vol. 8, no. 10, pp. 355-358, 2017.
[22] A.V.D. Oord, S. Dieleman, and B. Schrauwen, "Deep content-based music recommendation," 26th International Conference on Neural Information Processing Systems, vol. 2, pp. 2643-2651, 2013.
[23] M.A. Rahman, A. Anjum, M.M.H. Milu, F. Khanam, M.S. Uddin, and M.N. Mollah, "Emotion recognition from EEG-based relative power spectral topography using convolutional neural network," Array, vol. 11, 2021.
[24] L. Dubreuil-Vall, G. Ruffini, and J.A. Camprodon, "Deep Learning Convolutional Neural Networks Discriminate Adult ADHD From Healthy Individuals on the Basis of Event-Related Spectral EEG," Frontiers in Neuroscience, vol. 14, 2020.
[25] V.J. Lawhern, A.J. Solon, N.R. Waytowich, S.M. Gordon, C.P. Hung, and B.J. Lance, "EEGNet: A Compact Convolutional Network for EEG-based Brain-Computer Interfaces," Journal of Neural Engineering, vol. 15, no. 5, 2018.
[26] S. Xingjian, Z. Chen, H. Wang, D.Y. Yeung, W.K. Wong, and W.C. Woo, "Convolutional LSTM network: A machine learning approach for precipitation nowcasting," Advances in neural information processing systems, pp. 802-810, 2015.
[27] R. Zhao, W. Ouyang, H. Li, and X. Wang, "Saliency detection by multi-context deep learning," IEEE Conference on Computer Vision and Pattern Recognition, pp. 1265-1274, 2015.
[28] J. Yosinski, J. Clune, A. Nguyen, T. Fuchs, and H. Lipson, "Understanding neural networks through deep visualization," International Conference on Machine Learning, pp. 6-11, 2015.
[29] M.D. Zeiler, and R. Fergus, "Visualizing and understanding convolutional networks," European conference on computer vision, pp. 818-833, 2014.
[30] S. Sarkar, V. Venugopalan, K. Reddy, J. Ryde, M. Giering, and N. Jaitly, "Occlusion edge detection in RGB-D frames using deep convolutional networks," IEEE High Performance Extreme Computing Conference, 2015.
[31] A. Mordvintsev, C. Olah, and M. Tyka, "DeepDream - a Code Example for Visualizing Neural Networks," Google Research Blog, 2015.
[32] R.R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, "Grad-CAM: Visual explanations from deep networks via gradient-based localization," IEEE International Conference on Computer Vision, pp. 618-626, 2017.
[33] B. Zhou, A. Khosla, A. Lapedriza, A. Oliva, and A. Torralba, "Learning deep features for discriminative localization," IEEE Conference on Computer Vision and Pattern Recognition, pp. 2921–2929, 2016.
[34] D.P. Kingma, and J. Ba, "Adam: A Method for Stochastic Optimization," International Conference on Learning Representations, 2015.
[35] M. Ahmadi, K. Kazemi, K. Kuc, A. Cybulska-Klosowicz, M. Zakrzewska, E. Racicka-Pawlukiewicz, M.S. Helfroush, and A. Aarabi, "Cortical source analysis of resting state EEG data in children with attention deficit hyperactivity disorder," Clinical Neurophysiology, vol. 131, no. 9, pp. 2115-2130, 2020.
[36] R. Aldemir, E. Demirci, H. Per, M. Canpolat, S. Özmen, and M. Tokmakçı, "Investigation of attention deficit hyperactivity disorder (ADHD) subtypes in children via EEG frequency domain analysis," International Journal of Neuroscience, vol. 128, no. 4, pp. 349-360. 2017.
[37] A.R. Clarke, R.J. Barry, R. McCarthy, and M. Selikowitz, "EEG analysis in Attention-Deficit/Hyperactivity Disorder: a comparative study of two subtypes," Psychiatry Research, vol. 81, no. 1, pp. 19-29, 1998.
[38] G. Ogrim, J. Kropotov, and K. Hestad, "The quantitative EEG theta/beta ratio in attention deficit/hyperactivity disorder and normal controls: sensitivity, specificity, and behavioral correlates," Psychiatry Research, vol. 198, no. 3, pp. 482-488, 2012.
[39] S.K. Loo, J.J. McGough, J.T. McCracken, and S.L. Smalley, "Parsing heterogeneity in attention-deficit hyperactivity disorder using EEG-based subgroups," Journal of Child Psychology and Psychiatry, vol. 59, no. 3, pp. 223-231, 2018.
[40] S. Markovska-Simoska, and N. Pop-Jordanova, "Quantitative EEG in Children and Adults With Attention Deficit Hyperactivity Disorder: Comparison of Absolute and Relative Power Spectra and Theta/Beta Ratio," EEG and Clinical Neuroscience Society, vol. 48, no. 1, pp. 20-32, 2017.
[41] B. Abibullaev, and J. An, "Decision support algorithm for diagnosis of ADHD using electroencephalograms," Journal of Medical Systems, vol. 36, no. 4, pp. 2675-2688, 2012.
[42] Z.H. Chang, "Development of a K-CPT-Based Hierarchy Inference System Utilizing Wireless EEG to Evaluate ADHD Symptoms in Preschoolers," National Yang Ming Chiao Tung University, 2021.
[43] H. Chen, Y. Song, and X. Li, "Use of deep learning to detect personalized spatial-frequency abnormalities in EEGs of children with ADHD," Journal of Neural Engineering, vol. 16, no. 6, 2019.
[44] L. Dubreuil-Vall, G. Ruffini, and J. Camprodon, "A deep learning approach with event-related spectral EEG data in attentional deficit hyperactivity disorder," medRxiv, 2019.
[45] P.T. Boer, D.P. Kroese, S. Mannor, and R.Y. Rubinstein, "A tutorial on the cross-entropy method," Annals of Operations Research, vol. 134, no. 1, pp. 19-67, 2005.
[46] M.A. Hearst, S.T. Dumais, E. Osuna, J. Platt, and B. Scholkopf, "Support vector machines," IEEE Intelligent Systems and their Applications, vol. 13, no. 4, pp. 18-28, 1998.