Multi-Site Diagnostic Classification of Schizophrenia Using Discriminant Deep Learning with Functional Connectivity...

Area of Science:

• Neuroscience
• Psychiatry
• Artificial Intelligence

Background:

• Single-site brain imaging studies lack generalizability for diagnosing psychiatric disorders like schizophrenia.
• Advanced deep learning (DL) offers potential for cross-site classification by learning subtle patterns and overcoming site variations.
• DL-based cross-site transfer classification remains unexplored for schizophrenia diagnosis.

Purpose of the Study:

• To investigate the efficacy of deep learning for cross-site classification of schizophrenia using functional MRI data.
• To develop a DL model capable of learning site-shared functional connectivity features for schizophrenia diagnosis.
• To assess the generalizability and accuracy of DL models across multiple independent imaging sites.

Main Methods:

• Collected a large multi-site functional MRI dataset (n=734) with 357 individuals diagnosed with schizophrenia.
• Developed a deep discriminant autoencoder network to identify functional connectivity patterns.
• The model was trained to discriminate between schizophrenic patients and healthy controls, focusing on site-shared features.

Main Results:

• Achieved approximately 85.0% accuracy in multi-site pooling classification and 81.0% in leave-site-out transfer classification.
• Identified dysregulation within the cortical-striatal-cerebellar circuit in schizophrenia.
• Discriminating functional connections were predominantly located within and across the default, salience, and control networks.

Conclusions:

• Dysfunctional integration of the cortical-striatal-cerebellar circuit across key brain networks may underlie schizophrenia's pathophysiology.
• The proposed DL method demonstrates potential for learning reliable brain connectome patterns.
• This approach may improve schizophrenia prediction and understanding across diverse imaging sites.