MAMF-GCN model for anxious and non-anxious depression classification and neuroimaging marker recognition

Area of Science:

• Neuroscience
• Artificial Intelligence
• Psychiatry

Background:

• Depression is a widespread mental health disorder.
• Differentiating between anxious and non-anxious depression presents diagnostic challenges.
• Identifying reliable neuroimaging markers for depression subtypes is crucial.

Purpose of the Study:

• To develop and validate a deep learning model for classifying anxious and non-anxious depression.
• To identify neuroimaging markers that distinguish between these depression subtypes.
• To leverage both imaging and non-imaging data for improved classification accuracy.

Main Methods:

• Utilized REST-meta-MDD public dataset (158 anxious, 108 non-anxious depression patients).
• Computed neuroimaging metrics: fALFF, DC, ReHo, VMHC, and FC.
• Developed a multi-scale adaptive multi-channel fusion deep graph convolutional network (MAMF-GCN) integrating clinical data (site, age, gender, HAMD).
• Applied gradient backpropagation for feature screening and marker identification.

Main Results:

• The MAMF-GCN model achieved high classification performance: 98.12% accuracy, 99.41% sensitivity, 97.46% specificity, 98.38% F1-score, and 0.9867 AUC.
• Demonstrated superior performance compared to state-of-the-art methods.
• Confirmed the effectiveness of integrating non-imaging clinical data for enhanced classification.
• Identified the cerebellar network, visual, auditory, and motor control networks as critical for differentiating depression subtypes.

Conclusions:

• The proposed method is highly effective for classifying anxious and non-anxious depression.
• Successfully identified potential neuroimaging markers for depression subtyping.
• The approach holds significant clinical value for diagnosing and potentially treating depression subtypes.