機器學習和 EEG 信號視覺化預測重度憂鬱症患者的經顱磁刺激治療反應

重度憂鬱症被認為是一種導致認知和情感障礙的慢性疾病，且具有易惡化和與其他症狀高度合併風險的特性。重度憂鬱症在世界上相當普及，但其中一定比例的患者卻未能在常用的抗憂鬱藥物治療方式下好轉。對於此類患者，重複性經顱磁刺激有機會產生療效。本研究針對重度憂鬱症患者的腦電圖訊號進行分析，將其分解為 delta，theta，alpha，beta 和 gamma 五個頻段，應用數種線性和非線性法提取不同面向的特徵以機器學習預測重複性經顱磁刺激對該病患的療效。而因應腦電訊號圖並不具有客觀的標準波型和數值，本研究利用視覺化方式呈現不同頻段的腦電訊號圖變化對照醫學理論以驗證機器學習的預測果。此外，分析過程中包含各通道腦電訊號圖的前處理結果、各通道與特定區域在不同頻段下之腦電訊號圖數值，以及腦電訊號圖中疑似含有偽影成分的片段標記也將一並呈現以供醫師做為最終判讀的參考。實驗結果顯示經由機器學習區分重複性經顱刺激治療有效及無效者可達到 90%的準確率，其中 TPR 為 85.7%，TNR 為 100%。而被區分為治療有效者相比於治療無效者在前額葉 theta 頻段及 delta 頻段的腦電訊號圖視覺化結果亦可見顯著變化，證實了機器學習預測結果的可信度。

關鍵字

重度憂鬱症；腦電訊號圖；重複性經顱磁刺激；機器學習；視覺化

並列摘要

Major Depressive Disorder (MDD) is considered as a chronic disease that causes cognitive and affective impairments, and has the characteristics of easy deterioration and high comorbidity. MDD is quite common in the world, but a significant proportion of patients fail to response to commonly used antidepressant. For such patients, repetitive transcranial magnetic stimulation (rTMS) may be effective. In this study, we analyze the electroencephalograph (EEG) signals of MDD patients, decomposed them into five frequency bands, delta theta, alpha, beta, and gamma, and applied several linear and nonlinear methods to extract features for machine learning to predict the effect of rTMS on the patient. Since the EEG signal does not have objective standard waveforms and values, visualization is used in this study to present the changes of the EEG signals in different frequency bands and compares medical theories to verify the prediction results of machine learning. In addition, some information including the preprocessing results of the EEG signal of each channel, the EEG signal values of each channel and specific area in different frequency bands, and the segments of the EEG signals that are marked as suspected of containing artifacts will also be present for physician as a reference for the final judgement. The experimental results show that the machine learning for distinguishing responders and non-responders can achieve a validation accuracy of 90%, with TPR of 85.7% and TNR of 100%. The visualization results of the EEG signals in the frontal theta band and frontal delta band can also show significant changes between those who are classified as responders and those who are classified as non-responders, which confirms the prediction results of machine learning.

並列關鍵字

Major Depressive Disorder ； Electroencephalograph ； Repetitive transcranial magnetic stimulation ； Machine learning ； Visualization

參考文獻

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

Google Scholar

[2] World Health Organization (WHO). (2017). Depression: Fact sheet. Available: http://www.who.int/mediacentre/factsheets/fs369/en/

Google Scholar

[3] The Washington Post, “ A stunning map of depression rates around the world, “

Google Scholar

Available: https://www.washingtonpost.com/news/worldviews/wp/2013/11/07/a stunning-map-of-depression-rates-around-the-world/?noredirect=on

Google Scholar

[4] Ministry of Health and Welfare and National Health Insurance Administration,Available: https://www.nhi.gov.tw/index2015.aspx, 2016

Google Scholar

國際替代計量

機器學習和 EEG 信號視覺化預測重度憂鬱症患者的經顱磁刺激治療反應

未授權

主題瀏覽