A unified framework for MR based disease classification.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Machine Learning

Background:

• Structural brain differences are observed in schizophrenia.
• Accurate classification of schizophrenia using MRI is crucial for diagnosis and research.
• Existing methods for analyzing brain structure in schizophrenia have limitations.

Purpose of the Study:

• To develop and validate a novel method for detecting structural brain differences between schizophrenia patients and healthy controls using anatomical parameterization of spatial warps.
• To classify individuals into clinical groups (schizophrenia vs. healthy control) based on MRI data.
• To interpret anatomical differences by visualizing discriminative warps.

Main Methods:

• Employed structure-specific 9-parameter affine transformations to represent spatial warps for global, non-rigid mapping.
• Estimated transformation parameters by minimizing Kullback-Leibler divergence directly from medical scans.
• Utilized a linear Support Vector Machine classifier with estimated parameters for group assignment.

Main Results:

• Achieved up to 90% leave-one-out cross-validation accuracy in classifying first-episode schizophrenia patients.
• Demonstrated favorable comparison with state-of-the-art techniques in schizophrenia MRI research.
• Enabled visualization of discriminative warps, aiding in the interpretation of anatomical differences.

Conclusions:

• The proposed anatomical parameterization of spatial warps is an effective method for structural analysis and classification in schizophrenia.
• The approach provides a direct, accurate, and interpretable way to identify disease-related brain changes from MRI.
• This technique holds promise for advancing schizophrenia research and potentially clinical applications.