Related Experiment Video
Updated: Jul 8, 2025
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
Multi feature fusion network for schizophrenia classification and abnormal brain network recognition.
Chang Wang1, Chen Wang1, Yaning Ren1
1The Second Affiliated Hospital of Xinxiang Medical University, Xinxiang, China; Henan Key Laboratory of Biological Psychiatry, Xinxiang, China; School of Medical Engineering, Xinxiang Medical University, Xinxiang, China; Engineering Technology Research Center of Neurosense and Control of Henan Province, Xinxiang, China; Xinxiang Engineering Technology Research Center of Intelligent Medical Imaging Diagnosis, Xinxiang, China.
This study introduces a multi-feature fusion network (MFFN) for schizophrenia classification, combining functional network connectivity (FNC) and time courses (TC). The MFFN effectively identifies schizophrenia patients with high accuracy, highlighting aberrant brain network connections.
Area of Science:
- Neuroscience
- Computational Psychiatry
- Machine Learning in Healthcare
Background:
- Schizophrenia classification and abnormal brain network recognition are crucial research areas.
- Existing machine learning and deep learning methods for schizophrenia classification have limitations.
- Few studies leverage multi-feature fusion to enhance schizophrenia representation.
Purpose of the Study:
- To propose a multi-feature fusion network (MFFN) for distinguishing schizophrenia patients from healthy controls.
- To integrate functional network connectivity (FNC) and time courses (TC) for improved schizophrenia detection.
- To identify discriminative brain networks associated with schizophrenia.
Main Methods:
- Developed a multi-feature fusion network (MFFN) integrating deep neural network (DNN) and C-RNNAM backbones.
- Utilized Deep SHAP to identify the most discriminative brain networks.
- Validated the model on two public datasets using quantitative evaluation metrics.
Main Results:
- The MFFN achieved high classification accuracy (ACC=87.30%, AUC=0.9081), outperforming state-of-the-art methods.
- Functional network connectivity (FNC) derived from independent component analysis showed an advantage over static and dynamic functional connections.
- Ablation experiments confirmed the benefits of multi-feature fusion and attention mechanisms.
- Identified aberrant connections in the default mode and visual networks in schizophrenia patients.
Conclusions:
- The proposed MFFN effectively identifies schizophrenia and visualizes abnormal brain networks.
- The method demonstrates significant clinical application value for schizophrenia diagnosis.
- Multi-feature fusion enhances the accuracy and interpretability of schizophrenia classification models.
