Multisite generalizability of schizophrenia diagnosis classification based on functional brain connectivity.

Area of Science:

• Neuroimaging
• Psychiatric Disorders
• Machine Learning

Background:

• Schizophrenia diagnosis relies on clinical symptoms, but neuroimaging offers objective biomarkers.
• Functional brain connectivity (FC) shows promise for classifying schizophrenia using fMRI data.
• Assessing the generalizability of FC-based classifiers across diverse settings is crucial for clinical translation.

Purpose of the Study:

• To evaluate the generalizability of schizophrenia classification based on functional brain connectivity.
• To determine the impact of training data origin (single vs. multiple sites) on classifier performance.
• To assess generalizability across different scanning sites and cognitive task conditions.

Main Methods:

• Utilized functional magnetic resonance imaging (fMRI) data from 191 schizophrenia patients and 191 healthy controls.
• Data were acquired from 6 different scanning sites under various task conditions.
• Tested classifier performance using different training-test scenarios, including single-site and multisite training datasets.

Main Results:

• Schizophrenia classification accuracy generalized well to new sites and cognitive contexts when classifiers were trained on multisite data.
• Training classifiers with data from a single site resulted in significantly lower classification accuracy.
• Multisite training demonstrated superior robustness and generalizability for fMRI-based diagnostic classifiers.

Conclusions:

• Training fMRI-based schizophrenia classifiers with multisite data is essential for achieving robust generalizability.
• Multisite training approaches are recommended for developing reliable diagnostic tools for widespread clinical use.
• Functional brain connectivity patterns, when trained on diverse datasets, can effectively aid in schizophrenia diagnosis.