The ability of fMRI to detect meaningful signals is limited by a number of factors that add error to each measurement. Some of these factors include thermal noise, system noise in the scanner, physiological noise from the subject, non-task related cognitive processes, and changes in cognitive strategy over time (Huettel et al., 2008; Kruger and Glover, 2001). The concept of reliability is, at its core, a representation of the ability to routinely detect relevant signals from this background of meaningless noise. If a voxel timeseries contains a large amount of signal then the primary sources of variability are actual changes in blood flow related to neural activity within the brain. Conversely, in a voxel containing a large amount of noise the measurements are dominated by error and would not contain meaningful information. By increasing the amount of signal, or decreasing the amount of noise, a researcher can effectively increase the quality and reliability of acquired data.
The quality of data in magnetic resonance imaging is typically measured using the signal-to-noise ratio (SNR) of the acquired images. The goal is to maximize this ratio. Two kinds of SNR are important for functional MRI. The first is the image SNR. It is related to the quality of data acquired in a single fMRI volume. Image SNR is typically computed as the mean signal value of all voxels divided by the standard deviation of all voxels in a single image:
Increasing the image SNR will improve the quality of data at a single point in time. However, most important for functional neuroimaging is the amount of signal present in the data across time. This makes the temporal SNR (tSNR) perhaps the most important metric of data for functional MRI. It represents the signal-to-noise ratio of the timeseries at each voxel:
The tSNR is not the same across all voxels in the brain. Some regions will have higher or lower tSNR depending on location and constitution. For example, there are documented differences in tSNR between gray matter and white matter (Bodurka et al., 2005). The typical tSNR of fMRI can also vary depending on the same factors that influence image SNR.
Another metric of data quality is the contrast-to-noise ratio (CNR). This refers to the ability to maximize differences between signal intensity in different areas in an image (image CNR) or to maximize differences between different points in time (temporal CNR). With regard to functional neuroimaging, the temporal CNR represents the maximum relative difference in signal intensity that is represented within a single voxel. In a voxel with low CNR there would be very little difference between two conditions of interest. Conversely, in a voxel with high CNR there would be relatively large differences between two conditions of interest. The image CNR is not critical to fMRI, but having a high temporal CNR is very important for detecting task effects.
It is generally accepted that fMRI is a rather noisy measurement with a characteristically low tSNR, requiring extensive signal averaging to achieve effective signal detection (Murphy et al., 2007). The following sections provide greater detail on the influence of specific factors on the SNR/tSNR of functional MRI data. We break these factors down by the influence of differences in image acquisition, the image analysis pipeline, and the contribution of the subjects themselves.
No comments:
Post a Comment