基於機器學習架構的經顱磁刺激抗鬱療效預測於重度憂鬱症之腦波即時分析

Real Time EEG Analysis for Prediction of Antidepressant Responses of Transcranial Magnetic Stimulation in Major Depressive Disorder Based on Machine Learning

李易宸(Yi-Chen Li)

指導教授 ： 陳中平 ； 孫啟光

共同指導教授 ： 李正達(Cheng-Ta Li)

國立臺灣大學/電機資訊學院/生醫電子與資訊學研究所/碩士(2019年)

https://doi.org/10.6342/NTU201902399

摘要

重度憂鬱症被認為是一種傾向慢性，病程易惡化的疾病，具有與其他症狀高度合併風險。具有一定比例的重度憂鬱症病患在數個抗憂鬱藥物的治療下沒有好轉，而這類型的病患卻有機會被重複性經顱磁刺激或間歇性脈衝式經顱磁刺激所治療。本研究分析了來自90位重度憂鬱症患者的腦電圖信號，分別觀察重複性經顱磁刺激或間歇性脈衝式經顱磁刺激治療之重度憂鬱症患者的特徵來預測其抗憂鬱反應。腦電信號被分解為5個頻段：delta，theta，alpha，beta和gamma頻段。分頻段之後我們應用數個非線性方法和線性方法做特徵提取，從腦電信號中提取不同面向的特徵。在結果中，首先我們確認了在增加收案數量後，前額葉theta頻段能夠區分重複性經顱磁刺激的治療有效者與無效者；接著我們發現前額葉delta頻段以及全頻段也能夠區分重複性經顱磁刺激的治療有效者與無效者；最後我們探索性的發現間歇性脈衝式經顱磁刺激的抗憂鬱反應可能與前額葉beta頻段有關聯。我們使用機器學習與深度學習用於區分重複性經顱磁刺激的治療有效以及無效者，validation accuracy達到91.1%，其中TPR為83.3%，FPR為5%。

關鍵字

重度憂鬱症 ； 腦電圖 ； 重複性經顱磁刺激 ； 間歇性脈衝式經顱磁刺激 ； 機器學習 ； 深度學習

並列摘要

Major depressive disorder (MDD) is increasingly recognized as a chronic, deteriorating illness with high comorbidity. A significant proportion of patients with MDD fail to respond to sequential antidepressants. Such treatment-resistant depression can be treated with noninvasive brain stimulation, such as repetitive transcranial magnetic stimulation (rTMS) and intermittent theta-burst stimulation (iTBS). In this study, we analyzed electroencephalograph (EEG) signals from a total of 90 patients with MDD. Antidepressant responses were predicted by observing the features of patients with MDD receiving rTMS or iTBS treatments. EEG signals were decomposed into 5 bands: delta, theta, alpha, beta, and gamma. Feature extraction, including linear and nonlinear methods, was applied to the EEG signals. First, our study demonstrated that frontal theta could be associated with antidepressant responses to rTMS instead of iTBS and sham treatments. Second, frontal delta and all bands (1–60 Hz) were associated with antidepressant responses to rTMS treatment. Third, antidepressant responses to iTBS treatment might be associated with frontal beta waves. Finally, machine learning and deep learning were used for distinguishing between responders and non-responders, and a 91% validation accuracy rate, an 83.3% true positive rate, and a 5.0% false positive rate were achieved.

並列關鍵字

Major depressive disorder ； Electroencephalograph ； Repetitive transcranial magnetic stimulation ； Intermittent theta-burst stimulation ； Machine learning ； Deep learning

參考文獻

[1] K. R. Krishnan, "Comorbidity and depression treatment, " Biological psychiatry, vol. 53, pp. 701-706, 2003.

[2] B. Kennard, S. Silva, B. Vitiello, J. Curry, C. Kratochvil, A. Simons, et al., "Remission and residual symptoms after short-term treatment in the Treatment of Adolescents with Depression Study (TADS), " Journal of the American Academy of Child and Adolescent Psychiatry, vol. 45, pp. 1404-1411, 2006.

[3] N. Kennedy, E.S. Paykel, "Residual symptoms at remission from depression: impact on long-term outcome, " Journal of affective disorders, vol. 80, pp. 135-144, 2004.

[4] S.P. Roose, A.H. Glassman, S.N. Seidman, "Relationship between depression and other medical illnesses, " JAMA : the journal of the American Medical Association, vol. 286, pp. 1687-1690, 2001.

[5] D. A. Pizzagalli, "Frontocingulate dysfunction in depression: Toward biomarkers of treatment response," Neuropsychopharmacology, vol. 36, pp. 183-206, 2011.