Statistical Parametric Mapping

Statistical Parametric Mapping
Name	Statistical Parametric Mapping
Developer	Wellcome Trust Centre for Neuroimaging
Released	1991
Operating system	Cross-platform
Genre	Neuroimaging analysis
License	GNU General Public License

Statistical Parametric Mapping. It is a cornerstone methodology for analyzing functional neuroimaging data, most prominently from functional magnetic resonance imaging and positron emission tomography. The technique constructs spatially extended statistical processes to test hypotheses about regionally specific effects within brain mapping studies. Developed primarily at the Wellcome Trust Centre for Neuroimaging at University College London, it has become a standard framework in cognitive neuroscience and clinical neuroscience.

Overview

This approach provides a unified framework for the statistical analysis of multivariate brain imaging data sets. The core concept involves creating a map of test statistics, where each voxel in the brain image is assigned a value based on a general linear model. These maps are then assessed for statistical significance, allowing researchers to identify brain regions where activity is correlated with an experimental task or condition. The methodology integrates principles from Gaussian random field theory to handle the problem of multiple comparisons inherent in analyzing thousands of voxels simultaneously.

Methodological Foundations

The methodological bedrock rests on the application of the general linear model to each voxel's time series. This model expresses the observed blood-oxygen-level dependent signal as a linear combination of explanatory variables, plus an error term. Key preprocessing steps, including spatial normalization to a standard space like the Montreal Neurological Institute template, spatial smoothing with a Gaussian kernel, and temporal filtering, are essential. These steps ensure data conform to the model's assumptions and improve the validity of subsequent statistical inference. The work of pioneers like Karl J. Friston was instrumental in formalizing these foundations.

Statistical Inference in SPM

Inference moves from the single voxel to the cluster or set level using Gaussian random field theory. This theory provides approximations for the probability of obtaining clusters of activated voxels of a given size under the null hypothesis. The process involves setting a primary threshold for the test statistic and then assessing the significance of resulting clusters based on their spatial extent. Alternatives like false discovery rate control are also implemented. This framework addresses the multiple comparisons problem more powerfully than simplistic Bonferroni correction, by leveraging the spatial correlation inherent in imaging data.

Applications in Neuroimaging

Its primary application is the analysis of data from functional magnetic resonance imaging studies in cognitive psychology and psychiatry. It is extensively used to localize brain regions involved in processes like memory, attention, language, and emotion. Beyond basic research, it is applied in clinical populations to study disorders such as Alzheimer's disease, schizophrenia, and major depressive disorder. The technique is also fundamental to psychopharmacology studies using positron emission tomography to measure neuroreceptor binding and has been adapted for other modalities like electroencephalography and magnetoencephalography.

Software Implementation

The methodology is implemented in a software package written in MATLAB, which is distributed freely under the GNU General Public License. The software suite, developed and maintained by the Wellcome Trust Centre for Neuroimaging, provides a comprehensive environment for data preprocessing, model specification, estimation, and inference. It is designed to interface seamlessly with other neuroimaging tools and standard image formats. The software's development has been closely tied to the methodological advances published in journals like NeuroImage and Human Brain Mapping.

Limitations and Criticisms

Criticisms often focus on the assumptions underlying Gaussian random field theory, which may not hold with insufficient spatial smoothing or in subcortical brain regions. The reliance on mass-univariate analysis at each voxel has been contrasted with multivariate techniques like multivoxel pattern analysis that can detect distributed patterns of activity. Some argue that the standard general linear model approach may not fully capture the hemodynamic response's complexity. Furthermore, concerns about the reproducibility of neuroimaging findings and appropriate statistical power have prompted ongoing methodological refinements within the framework.

Category:Neuroimaging Category:Biomedical cybernetics Category:Computational neuroscience