Classical Statistics and Statistical Learning in Imaging Neuroscience.

Area of Science:

• Neuroimaging
• Statistical Analysis
• Machine Learning

Background:

• Classical statistical methods like t-tests and ANOVA have long been standard in brain imaging.
• Statistical learning (SL) methods are increasingly used for complex neuroimaging datasets.
• A need exists to differentiate between classical and SL approaches in neuroimaging data analysis.

Purpose of the Study:

• To discuss the implications of inferential justifications and algorithmic methodologies in neuroimaging data analysis.
• To clarify the distinctions between classical statistics and statistical learning in the context of brain imaging.
• To reduce confusion in selecting appropriate analytical techniques for neuroimaging research.

Main Methods:

• Conceptual paper reviewing statistical methodologies in neuroimaging.
• Comparison of classical statistical approaches (e.g., regression, t-tests, ANOVA) with statistical learning methods (e.g., cross-validation, pattern classification, sparsity-inducing regression).
• Analysis of theoretical foundations, assumptions, and outcome metrics of both statistical paradigms.

Main Results:

• Classical statistics and statistical learning stem from different origins and theoretical bases.
• Each statistical approach makes distinct assumptions and employs different evaluation metrics.
• The two methodologies yield differently nuanced conclusions for neuroimaging data.

Conclusions:

• Classical statistics and statistical learning offer complementary but distinct insights in neuroimaging.
• Understanding the foundational differences is crucial for appropriate application and interpretation.
• Clarifying these distinctions aids researchers in selecting optimal methods for brain imaging analysis.