Differential Diagnosis of Attention-Deficit / Hyperactivity Disorder and Bipolar Disorder using Steady-State Visual Evoked Potentials

Bipolar disorder and Attention-deficit/Hyperactivity disorder (ADHD) are two prevalent disorders whose symptoms are similar. In order to reduce the misdiagnosis between bipolar disorder and ADHD, a machine learning-based system using electroencephalography (EEG) and steady state potentials (i.e., steady-state visual evoked potential [SSVEP]) was evaluated to classify ADHD, bipolar disorder and normal conditions. Indeed, this research was conducted for the first time with the aim of designing a machine learning system for EEG detection of ADHD, bipolar disorder, and normal conditions using SSVEPs. For this purpose, both linear and nonlinear dynamics of extracted SSVEPs were analyzed. Indeed, after data preprocessing, spectral analysis and recurrence quantification analysis (RQA) were applied to SSVEPs. Then, feature selection was utilized through the DISR. Finally, we utilized various machine learning techniques to classify the linear and nonlinear features extracted from SSVEPs into three classes of ADHD, bipolar disorder and normal: k-nearest neighbors (KNN), support vector machine (SVM), linear discriminant analysis (LDA) and Naive Bayes. Experimental results showed that SVM classifier with linear kernel yielded an accuracy of 78.57% for ADHD, bipolar disorder and normal classification through the leave-one-subject-out (LOSO) cross-validation. Although this research is the first to evaluate the utilization of signal processing and machine learning approaches in SSVEP classification of these disorders, it has limitations that future studies should investigate to enhance the efficacy of proposed system.