Visualizing Functional Network Connectivity Difference between Healthy Control and Major Depressive Disorder Using an...

Area of Science:

• Neuroscience
• Computational Psychiatry
• Medical Imaging Analysis

Background:

• Major depressive disorder (MDD) is a prevalent mental health condition with complex neurobiological underpinnings.
• Previous research identified alterations in brain networks like the default mode and cognitive control networks in MDD patients compared to healthy controls (HC).
• Advanced machine learning techniques are increasingly applied to neuroimaging data for classifying psychiatric disorders, but model interpretability remains a challenge.

Purpose of the Study:

• To classify individuals with major depressive disorder (MDD) from healthy controls (HC) using whole-brain connectivity patterns.
• To interpret machine learning models used for MDD classification by identifying key differentiating features.
• To explore the role of various brain networks, beyond traditionally studied ones, in distinguishing MDD from HC.

Main Methods:

• Whole-brain connectivity was estimated and used to train multiple machine learning classifiers: support vector machine (SVM), random forest, XGBoost, and convolutional neural network (CNN).
• The SHapley Additive exPlanations (SHAP) approach was employed as a feature learning method to interpret the classification models and understand feature importance.
• Classification accuracy and feature importance were evaluated across all employed machine learning methods.

Main Results:

• All tested classification methods achieved consistent accuracy in distinguishing MDD from HC subjects.
• SHAP analysis successfully identified key brain connectivity features contributing to the classification, providing model interpretability.
• The study highlighted the significant involvement of visual and sensory-motor networks, in addition to default mode and cognitive control networks, in differentiating between MDD and HC.

Conclusions:

• Machine learning models, particularly when interpreted with SHAP, can effectively classify MDD from HC based on whole-brain connectivity.
• Brain connectivity patterns within visual and sensory-motor networks are important biomarkers for MDD.
• This approach offers a promising avenue for understanding the neurobiological basis of depression and developing objective diagnostic tools.