8.4. Difference in meanings between different toolboxes

Page summary

• SPM has the same meaning of analysis levels, compared to Nilearn’s models

  • First level: analyze across runs

  • Second level: group level analysis

• FSL has a different meaning

  • First level: analyze one run for one subject

  • Second level: analyze across runs for one subject

  • Third level: group level analysis

These differences can be put in the table below:

Analyze	nilearn	SPM	FSL
One run, one subject	First level	First level	First level
More than one run, one subject	First level	First level	Second level
More than or equal to one run, more than one subject	Second level	Second level	Third level

Table showing the differences of the meaning for level of models between toolboxes/libraries

8.4.1. Statistical Parametric Mapping (SPM)

SPM uses the same notation as Nilearn for analysis levels, with a note that in SPM terminology session refers to an imaging run, and within a run there could be multiple conditions (for example congruent and incongruent). SPM provides tutorials and documentation, including lectures, to help users analyze their own fMRI data with the meaning of analysis levels being as follows:

• First level analysis in SPM: Analyze across runs for a subject (meaning more than one run of one subject)

• Second level analysis in SPM: Analyze across several subjects (meaning more than one subject with one or more run per subject). This is also known as group level analysis which test if the average estimate across subjects is statistically significant.

Fixed effects analyses

One important difference between SPM and Nilearn, is that the typical first level workflow in SPM will create a single design matrix for all runs and thus run a single model at the subject level (see First level analysis in SPM). Nilearn will instead create one design matrix per run, and run one model per run (see for example this report with 2 runs). In Nilearn, to compute summary statistics across runs you can use the method compute_contrast or the function compute_fixed_effects.

See also

Simple example of two-runs fMRI model fitting

8.4.2. FMRIB Software Library (FSL)

FSL uses a slightly different set of meanings for analysis levels, but with a note that a session still refers to an imaging run. Specifically, FEAT, FSL software tool for model-based fMRI data analysis, provides a CLI and GUI to run analysis on imaging data for first level and higher level analysis with the terminology meaning as follows:

• First level analysis in FSL: Analyze each run’s data by getting the parameter and contrast estimates (meaning one run of one subject)

• Second level analysis in FSL: Analyze across runs for a subject via averaging the parameter and contrast estimates within each subject (meaning more than one run of one subject)

• Third level analysis in FSL: Analyze across several subjects or group level analysis on the averaged contrast estimates for all subjects within the group (meaning more than one subject with one or more runs per subject)