Related Experiment Video
Updated: Jul 1, 2026
14:27
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
15.6K
STANet: A Novel Spatio-Temporal Aggregation Network for Depression Classification with Small and Unbalanced FMRI Data
Wei Zhang1, Weiming Zeng1, Hongyu Chen1
1Lab of Digital Image and Intelligent Computation, College of Information Engineering, Shanghai Maritime University, Shanghai 201306, China.
Tomography (Ann Arbor, Mich.)
|December 27, 2024
Summary
This study introduces STANet, a novel AI network for diagnosing depression using functional magnetic resonance imaging (fMRI) brain activity. STANet improves diagnostic accuracy by analyzing spatio-temporal features and balancing imbalanced datasets.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Medical Diagnostics
Background:
- Early depression diagnosis is vital for treatment and suicide prevention.
- Current diagnostic methods lack objective biomarkers.
- Functional magnetic resonance imaging (fMRI) combined with AI offers potential but faces challenges with small, imbalanced datasets.
Purpose of the Study:
- To develop an advanced AI model for accurate depression diagnosis using fMRI data.
- To address the limitations of small and imbalanced datasets in neuroimaging research.
- To improve the clinical utility of fMRI in depression assessment.
Main Methods:
- Proposed the Spatio-Temporal Aggregation Network (STANet), integrating CNNs and RNNs.
- Utilized Independent Component Analysis (ICA) for spatio-temporal aggregation and multi-scale deep convolution for feature extraction.
- Employed Synthetic Minority Over-sampling Technique (SMOTE) for data balancing and an Attention-Fourier Gate Recurrent Unit (AFGRU) classifier for enhanced dependency capture.
Main Results:
- STANet achieved 82.38% accuracy and 90.72% AUC in depression diagnosis.
- The Spatio-Temporal Feature Aggregation module and AFGRU classifier significantly improved performance.
- Spatio-temporal features outperformed purely temporal or spatial features, and SMOTE was effective for data balancing.
Conclusions:
- STANet demonstrates superior performance in diagnosing depression from fMRI data, outperforming existing methods.
- The model effectively handles small and imbalanced datasets, crucial for clinical applications.
- STANet offers a promising tool for enhancing depression diagnosis and treatment assessment in clinical settings.
