Single Channel EEG Classification: A Case Study on Prediction of Major Depressive Disorder Treatment Outcome

• Published in: IEEE access Vol. 9; pp. 3417 - 3427
• Main Authors: Hasanzadeh, Fatemeh, Mohebbi, Maryam, Rostami, Reza
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2021; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Case studies; Classification; Classifiers; Clinical outcomes; Datasets; Depression; Electrodes; Electroencephalography; Feature extraction; K-nearest neighbors algorithm; major depressive disorder; Mental depression; prediction treatment response; Redundancy; single channel EEG; Spatial resolution; Testing; Time series analysis; Training; Transcranial magnetic stimulation; Wearable technology
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2020.3046993

In multichannel EEG, several electrodes are attached to the head that may be annoying for patients and troublesome for operators. Moreover, the number of electrodes is the main reason of the infeasibility of developing EEG based wearable and point of care devices. To address this problem, recently, the concept of single-channel EEG (SCEEG) is presented. The spatial resolution of SCEEG is lower than the multichannel one, but it is easy to use, cost-effective, ubiquitous, and wearable. In this paper, for the first time, we have developed the concept of SCEEG for the classification of responders and nonresponders to repetitive transcranial magnetic stimulation (rTMS) treatment in major depressive disorder (MDD). We also compared the performance of SCEEG and multichannel EEG with the different number of channels in the prediction of responding to rTMS treatment. 19-electrode EEG is recorded from 46 MDD patients before rTMS treatment. Among participants, 23 individuals responded to treatment. The dataset is partitioned into the training (36 subjects) and testing (10 subjects) datasets. Linear and nonlinear features were extracted from every channel of EEG. In training, to select informative features, the minimal-redundancy-maximal-relevance (mRMR) algorithm was applied. The selected features were classified by k-nearest neighbors (KNN) classifier, which is evaluated by leave-one-out cross-validation. Then the obtained classifier is applied to the testing dataset. The results demonstrated that the F8 channel classifies responders and nonresponders with an accuracy of 80%. Moreover, our results revealed that SCEEG could perform as multichannel EEG in the prediction of rTMS treatment outcome in MDD patients. The obtained accuracy indicates that our proposed method based on SCEEG has a high potential for predicting rTMS treatment outcome in MDD patients.