Structural and diffusion MRI based schizophrenia classification using 2D pretrained and 3D naive Convolutional Neural...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Schizophrenia diagnosis relies on clinical symptoms, lacking objective biomarkers.
• Convolutional Neural Networks (CNNs) offer automatic feature learning for complex brain changes.
• Interpretable deep learning on multimodal neuroimaging for schizophrenia remains underexplored.

Purpose of the Study:

• To develop and evaluate deep learning models for schizophrenia classification using multimodal 3D MRI data.
• To compare the performance of naive 3D CNN, pre-trained 2D CNN, and traditional machine learning approaches.
• To identify critical brain regions for schizophrenia classification and provide neurobiological insights.

Main Methods:

• Developed naive 3D CNN and utilized pre-trained 2D CNN models.
• Integrated 3D structural and diffusion Magnetic Resonance Imaging (MRI) data.
• Compared deep learning models against a handcrafted feature-based Support Vector Machine (SVM) approach.

Main Results:

• Naive 3D CNN models demonstrated superior performance over pre-trained 2D CNN and SVM.
• Multimodal neuroimaging models outperformed single-modality models.
• Identified critical grey and white matter regions, supporting salience network and striatal dysfunction hypotheses.

Conclusions:

• CNNs can automatically integrate multimodal 3D brain imaging features for schizophrenia identification.
• The study provides neurobiological interpretations crucial for objective diagnostic tools.
• Deep learning approaches show promise for developing interpretable imaging-based biomarkers in psychiatry.