Modelling the data and not the images in FMRI¶
Important
Check out What is new!
What is it?¶
This is the first statistical software tool which implements the model based (MB) estimator for functional magnetic resonance imaging (FMRI) data models. It is a new and original method for the statistical analysis of FMRI of brain scans. MB estimation combines all preprocessing steps of standard approaches into one modelling step. Without altering the original 4D-image, the method results in smooth fits of the underlying parameter fields. More importantly, the method yields a trustworthy estimate of the uncertainty in BOLD effect estimation.
Why should I care?¶
Current approaches to the analysis of FMRI data apply various preprocessing steps to the original FMRI. These preprocessings lead to a general underestimation of residual variance in the downstream analysis. This negatively impacts the type I error of statistical tests and increases the risk for reporting false positive results.
In contrast to standard approaches, MB estimation yields sound statistical estimates for the uncertainty in BOLD effect estimation. The availability of these uncertainty fields allows to model FMRI studies by random effects meta regression models, acknowledging that individual subjects are random entities, and that the certainty at which the individual BOLD effect can be estimated from an FMRI varies across the brain and between subjects.
Note
Please note that this is not the MATLAB program fmristat nor is it a reimplementation of the software. fmristat (now also part of niak) remains the reference implementation of Keith Worsley’s methods. Instead, this software contains an implementation of Möbius’ MB estimator.
MB estimation also encourages to process and report BOLD effects in ATI units. In particular multicentre studies gain power by its use. For more details, have a look at the chapter Population inference: Create a sample for further analysis.
How to cite estimator and software¶
Reference publication for the MB estimator is:
Thomas W. D. Möbius (2018) Modelling the data and not the images in FMRI, ArXiv e-prints, arXiv:1809.07232
Thomas W. D. Möbius (2018) fmristats: Modelling the data and not the images in FMRI (0.1.0) [Computer program]. Available at http://fmristats.github.io/
Thank you for using and citing this software (BibTeX:
references.bib
).
Installation¶
An easy way to install fmristats is to first install Anaconda3 and then to use pip to install fmristats itself. Anaconda3 is a cross platform distribution for data analysis and scientific computing. pip is the “PyPA recommended tool for installing Python packages”.
After installation of Anaconda3 run:
conda install pip numba numpy scipy matplotlib \
scikit-learn scikit-image \
pandas statsmodels seaborn
pip install fmristats
That’s it. You should be good to go.
If you are planing to use the fmristats wrappers to FSL and ANTS, please make sure that the respective command line tools are within your PATH.
Tutorials¶
Tutorials discussing various parts of fmristats; expect this list to grow.
- Getting to know fmristats
- Analysing multiple subjects
- Subject analysis: Prune a statistics field from non-brain areas
- Subject analysis: Inference at a single coordinate
- Population inference: Create a sample for further analysis
- Population inference: Confirmation study
- Population inference: Exploratory study
- Population inference: How to adjust for covariates
- Thresholding
- Run the analysis on a server
- Some notes on speed