A major depressive disorder diagnosis approach based on EEG signals using dictionary learning and functional connectivity features

• Published in: Australasian physical & engineering sciences in medicine Vol. 45; no. 3; pp. 705 - 719
• Main Authors: Movahed, Reza Akbari, Jahromi, Gila Pirzad, Shahyad, Shima, Meftahi, Gholam Hossein
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 01.09.2022; Springer Nature B.V
• Subjects: Biological and Medical Physics; Biomedical and Life Sciences; Biomedical Engineering and Bioengineering; Biomedicine; Biophysics; Classification; Diagnosis; Dictionaries; Electroencephalography; Learning; Medical and Radiation Physics; Mental depression; Performance measurement; Scientific Paper; Functional connectivity; Depression; Electroencephalogram (EEG); Dictionary learning; Major depressive disorder (MDD); Machine learning
• ISSN: 2662-4729; 0158-9938; 2662-4737; 1879-5447
• DOI: 10.1007/s13246-022-01135-1

Major depressive disorder (MDD) as a psychiatric illness negatively affects the behavior and daily life of the patients.Therefore, the early MDD diagnosis can help to cure the patients more efficiently and prevent adverse effects, although its unclear manifestations make the early diagnosis challenging. Nowadays, many studies have proposed automatic early MDD diagnosis methods based on electroencephalogram (EEG) signals. This study also presents an automated EEG-based MDD diagnosis framework based on Dictionary learning (DL) approaches and functional connectivity features. Firstly, a feature space of MDD and healthy control (HC) participants were constructed via functional connectivity features.Next, DL-based classification approaches such as Label Consistent K-SVD (LC-KSVD) and Correlation-based Label Consistent K-SVD (CLC-KSVD) methods, were utilized to perform the classification task. A public dataset was used, consisting of EEG signals from 34 MDD patients and 30 HC subjects, to evaluate the proposed method. To validate the proposed method, 10-fold cross-validation technique with 100 iterations was employed, providing accuracy (AC), sensitivity (SE), specificity (SP), F1-score (F1), and false discovery rate (FDR) performance metrics. The results show that LC-KSVD2 and CLC-KSVD2 performed efficiently in classifying MDD and HC cases. The best classification performance was obtained by the LCKSVD2 method, with average AC of 99.0%, SE of 98.9%, SP of 99.2%, F1 of 99.0%, and FDR of 0.8%. According to the results, the proposed method provides an accurate performance and, therefore, it can be developed into a computer-aided diagnosis (CAD) tool for automatic MDD diagnosis.