Disease Definition for Schizophrenia by Functional Connectivity Using Radiomics Strategy

Area of Science:

• Neuroimaging
• Radiomics
• Machine Learning

Background:

• Schizophrenia (SZ) diagnosis requires specific biomarkers for neurobiological substrates.
• Neuroimaging studies suggest disrupted functional connectivity is key in SZ pathophysiology.

Purpose of the Study:

• To develop and validate a radiomics-based method for defining schizophrenia using resting-state functional connectivity.
• To assess the diagnostic accuracy and generalizability of this method across different datasets.

Main Methods:

• Utilized resting-state functional connectivity data from 108 first-episode schizophrenia patients and 121 healthy controls across two datasets.
• Applied Least Absolute Shrinkage and Selection Operator (LASSO) for feature selection.
• Trained and validated multivariate classifiers using machine learning for diagnosis.

Main Results:

• Intra-dataset validation achieved 87.09% accuracy, with replication in an independent dataset yielding 82.61%.
• Inter-dataset validation demonstrated strong performance with 83.15% training accuracy and 80.07% testing accuracy.
• The findings confirm the utility of functional connectivity features for SZ diagnosis.

Conclusions:

• A valid radiomics approach using functional connectivity can diagnose schizophrenia.
• This method facilitates objective, individualized diagnosis of SZ through quantitative biomarkers.