Loading...

EEG-Based Markers of Major Depressive Disorder in Reinforcement Learning

Aghajani Zadeh, Zahra | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57710 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Vosoughi Vahdat, Bijan; Karbalaei Aghajan, Hamid
  7. Abstract:
  8. Major Depressive Disorder (MDD) is a common and serious mental health disorder characterized not only by mood disturbances but also by deficits in cognitive functions and decision-making processes. This disorder can affect all aspects of an individual’s life, including their relationships with family, friends, and society. Recent electroencephalography (EEG) studies have demonstrated that certain neuropsychiatric disruptions lead to alterations in specific brain signal metrics, which can serve as markers of brain dysfunction. Many studies have explored traditional linear EEG metrics, such as frequency band power, asymmetry in frequency band activity, and event-related potential components, to differentiate between depressed and healthy individuals. These studies have revealed the potential of EEG metrics to provide objective markers for diagnosing depression. EEG is inherently a nonlinear and nonstationary signal. Consequently, nonlinear metrics, such as complexity measures, are more appropriate for its analysis compared to linear ones. Moreover, differentiating EEG features during tasks related to depression symptoms can yield more relevant markers than analyzing resting-state signals. Therefore, this study investigates nonlinear EEG metrics, including Lempel-Ziv complexity, Higuchi’s fractal dimension, and sample entropy, during a probabilistic reward task associated with reinforcement learning, to distinguish between depressed and healthy groups. The results of this study indicate that the depressed group exhibited higher values of nonlinear EEG metrics during the probabilistic reward task in the time intervals of stimulus presentation, stimulus selection (participant action), and feedback observation compared to the control group. In classification results, the best performance (74.28% accuracy and 74.28% sensitivity) was achieved using a single nonlinear feature from F7 electrode during the stimulus selection interval, based on the sample entropy metric. These findings likely reflect differences in brain signals during reinforcement learning processes between depressed and healthy groups
  9. Keywords:
  10. Major Depressive ; Electroencephalography ; Reinforcement Learning ; Nonlinear Metrics ; Brain Waves

 Digital Object List

 Bookmark

No TOC