Improving the predictive potential of diffusion MRI in schizophrenia using normative models-Towards subject-level...

Area of Science:

• Neuroimaging
• Psychiatric Disorders
• Biostatistics

Background:

• Diffusion MRI reveals group differences in white matter between schizophrenia patients and controls.
• These group-level abnormalities are often not detectable at the individual subject level.
• Normative modeling aims to identify individual-level deviations from healthy brain data.

Purpose of the Study:

• To evaluate if normative modeling enhances subject-level prediction of schizophrenia.
• To assess the predictive power of individual deviations from normative white matter models.
• To investigate the utility of combining multiple diffusion MRI measures for classification.

Main Methods:

• Utilized a large, harmonized diffusion MRI dataset (512 controls, 601 schizophrenia patients).
• Calculated age and sex-adjusted z-scores for standard (fractional anisotropy) and advanced (free-water) diffusion MRI measures in 18 white matter regions.
• Compared predictive performance of extreme deviations versus combined information from multiple measures.

Main Results:

• Group differences in z-scores showed larger effect sizes than raw diffusion MRI values (p < .001).
• Predictions based on summary z-score measures had low predictive power (AUC < 0.63).
• Combining information from multiple white matter tracts and imaging measures improved prediction (best AUC = 0.726).

Conclusions:

• Extreme deviations from normative models are not optimal for subject-level schizophrenia prediction.
• Integrating the full distribution of deviations across multiple diffusion MRI measures enhances classification accuracy.
• This approach holds potential for improving subject-level diagnosis in psychiatric research.