PhysioNoise is a tool for physiological noise correction of fMRI. The core model of this tool is based on the noise model from RETROICOR (Glover et al. 2000), including cardiac and respiration causing noise in fMRI image.
The main idea of PhysioNoise is to build a GLM with noise as main components. Noise components come from RETROICOR noise model and are used as regressors in GLM. The residual of GLM will be the corrected fMRI data for furthur use.
The pipeline of this tool is divided into three modules.
This module mainly focus on preparing a fMRI data for GLM analysis. The operations include skull stripping and motion correction.
-
motion correction
-
registration to structural MRI
-
Stripping skull
The pipeline of this step is shown in the figure below. After transfering image values to fload, the reference for motion correction is collected from the middle volume of fMRI. Applying motion correction to the raw data with the reference, which is also plotted by last step. The preprocess gives back the frameworks of workflow, the plots of motion correction and preprocessed data.
The interfaces from the pipeline.
-
Concern: Respiration noise is part of motion correction. Interesting respirtion noise in GLM might be partly removed by motion correction, which may influence the results for respiration components.
This module calculates regressors for the final GLM step. Since fMRI data is a sequence of volumn images of each scanning time, the regressors are respondingly calculated for each volumn. Based on the noise model from RETROICOR (Glover et al. 2000), regressors in PhysioNoise consist of cardiac and respiration regressors.
According to RETROICOR model, physiological noise components can be expressed by a low order Fourier series expanded in terms of cardiac and respiration phases.
The cardiac phase and respiration phase are calculated from physiological signals. For each scanning time, a series of regressors are generated for denoise according to the given orders of Fourier expansion.
The procedure of denoise is based on GLM(Generalized linear model). To find the regions that best explain the data with calculated noise regressors, a first level analysis of fMRI data is built. Ideally, the regions found by the analysis should be related to physiological noise regressors. And the residual of the GLM is regarded as the denoised data.
In this step, contrasts of regressors are built to show the related regions of each regressor. Diagrams of first level analysis and contrasts are saved from this step.
Eventually, a residual will be saved to the output directory. And the z-statistical values of each regressor are also saved with a threshold of 3.5, from which the related noise regions can be recognized.
The order used in RETROICOR model is 2. And the order of regressors are:
Eight obvious regions according to eight noise regressors are shown below.
