An MLP-based model for identifying qEEG in depression

Area of Science:

• Neuroscience
• Computational Biology
• Animal Models

Background:

• Manual analysis of electroencephalography (EEG) recordings for depression is challenging.
• Quantitative EEG (qEEG) offers a more objective measure.
• Depression models in animals are crucial for understanding the condition.

Purpose of the Study:

• To develop and validate a Multilayer Perceptron (MLP) model for differentiating qEEG in depressed versus control animal models.
• To assess the efficacy of using specific frequency bands versus individual frequencies as input features for the MLP.
• To compare the MLP's classification rules with existing clinical observations in depression.

Main Methods:

• Utilized qEEG data from control, exercised, and depressed rats, focusing on frequencies from 1 to 30 Hz.
• Trained an MLP model using 30 individual frequency inputs and subsequently with 3 aggregated frequency bands (slow, medium, fast).
• Evaluated the MLP's performance in distinguishing between depressed and normal qEEG patterns.

Main Results:

• The MLP model successfully differentiated between normal and depressed rats with over 80% accuracy.
• The model classified most exercised rats' qEEG as normal.
• Reducing input features from 30 frequencies to 3 bands yielded comparable classification performance.

Conclusions:

• MLP analysis of qEEG is a viable method for identifying depression in animal models.
• The model's classification rules align with clinical perspectives on depression.
• Feature reduction to frequency bands maintains diagnostic accuracy, simplifying the model.