Related Experiment Video
Updated: Aug 17, 2026
Best Current Practice for Obtaining High Quality EEG Data During Simultaneous fMRI
Published on: June 3, 2013
Detecting and adjusting for artifacts in fMRI time series data
Jörn Diedrichsen1, Reza Shadmehr
1Department of Biomedical Engineering, Laboratory for Computational Motor Control, Johns Hopkins University School of Medicine, Baltimore, 720 Rutland Ave, 416 Traylor Building, MD 21205-2195, USA. jdiedric@bme.jhu.edu
Abstract:
We present a new method to detect and adjust for noise and artifacts in functional MRI time series data. We note that the assumption of stationary variance, which is central to the theoretical treatment of fMRI time series data, is often violated in practice. Sporadic events such as eye, mouth, or arm movements can increase noise in a spatially global pattern throughout an image, leading to a non-stationary noise process. We derive a restricted maximum likelihood (ReML) algorithm that estimates the variance of the noise for each image in the time series. These variance parameters are then used to obtain a weighted least squares estimate of the regression parameters of a linear model. We apply this approach to a typical fMRI experiment with a block design and show that the noise estimates strongly vary across different images and that our method detects and appropriately weights images that are affected by artifacts. Furthermore, we show that the noise process has a global spatial distribution and that the variance increase is multiplicative rather than additive. The new algorithm results in significantly increased sensitivity in the ability to detect regions of activation. The new method may be particularly useful for studies that involve special populations (e.g., children or elderly) where sporadic, artifact-generating events are more likely.
Insights
This study introduces a new method to handle noise in functional MRI (fMRI) data, improving the detection of brain activity. The approach adjusts for artifacts, enhancing sensitivity in fMRI analysis.
Area of Science:
- Neuroscience
- Biomedical Engineering
- Signal Processing
Background:
- Functional MRI (fMRI) data analysis relies on assumptions of stationary variance, often unmet in practice.
- Sporadic artifacts like movement introduce non-stationary noise, impacting fMRI time series.
- Existing methods may not adequately address global, spatially distributed noise patterns.
Purpose of the Study:
- To develop and validate a novel method for detecting and correcting noise and artifacts in fMRI time series.
- To improve the sensitivity of fMRI analysis by accounting for non-stationary noise.
- To provide a robust approach for fMRI studies, especially those involving populations prone to artifacts.
Main Methods:
- Derived a restricted maximum likelihood (ReML) algorithm to estimate noise variance per image.
- Employed weighted least squares to estimate linear model parameters using estimated variance.
- Applied the method to fMRI block design experiments to assess noise characteristics.
Main Results:
- Demonstrated significant variation in noise estimates across fMRI images.
- Showed the method effectively detects and weights artifact-affected images.
- Confirmed noise process is global and multiplicative, not additive.
- Observed a significant increase in sensitivity for detecting activation regions.
Conclusions:
- The new ReML-based method effectively addresses non-stationary noise in fMRI data.
- This approach enhances the reliability and sensitivity of fMRI analysis.
- The method is particularly beneficial for studies with special populations prone to artifacts.
