Critical Perspectives in Statistical Analysis of Neuroimaging Data
Andrew Bender, Department of Epidemiology and Bio-Statistics, Michigan State University, USA
Neuroimaging research is inherently interdisciplinary, but divergent perspectives between disciplines can complicate the success of a collaborative project. Statistical modeling and analysis of neuroimaging data can strongly benefit from neuroscientists’ perspectives on 1) brain organization, appreciating the brain as a complex organ composed of multiple, integrated subsystems; and 2) limitations and assumptions associated with different types of neuroimaging data, including design and processing of neuroimaging study data and potential sources of systematic error and bias. For example, among neuroimaging investigators, analytic perspectives widely range from strictly confirmatory, theoretically-guided approaches to exploratory or ‘hypothesis-free’ analyses, and each has its own advantages and downsides. Whereas confirmatory approaches may lend to easier interpretation or translation, they may under-identify novel effects. In contrast, exploratory approaches may provide new insights into neural organization, but have greater potential for generating spurious findings. This presentation will 1) address goals and values in neuroimaging data, including common points of practical divergence, 2) describe specific challenges in statistical analysis of structural and functional neuroimaging data, 3) identify current needs for new directions in statistical analysis, and, 4) propose alternatives to common approaches that may circumvent some of these issues.