Measuring heterogeneity in normative models as the effective number of deviation patterns

Area of Science:

• Computational biology
• Statistical modeling
• Data science

Background:

• Normative modeling characterizes clinical cohort deviations from reference populations.
• Existing methods lack robust measures for cohort heterogeneity.

Purpose of the Study:

• To extend normative modeling with a quantitative measure of cohort heterogeneity.
• To introduce the effective number of deviation patterns as a heterogeneity metric.

Main Methods:

• Utilized the Representational Rényi Heterogeneity method to generalize diversity measurement.
• Applied Gaussian process models to neuroimaging data.
• Employed variational autoencoders on handwritten image datasets.

Main Results:

• Developed a novel heterogeneity measure based on the effective number of deviation patterns.
• Demonstrated that a lower effective number of deviation patterns correlates with systematic deviations from normative distributions.
• Validated findings across both synthetic/real neuroimaging and handwritten image data.

Conclusions:

• The effective number of deviation patterns provides a robust measure of heterogeneity in normative modeling.
• This approach enhances the characterization of deviations in clinical and other data types.
• The method shows broad applicability across diverse scientific domains.