Impact of Polynomial Detrending on Weisskoff Analysis for Radius of Correlation Estimation
Abstract
Purpose
Weisskoff analysis assesses temporal stability of echo-planar imaging (EPI) in functional MRI (fMRI) by comparing how the temporal coefficient of variation (CoV) scales with ROI size relative to uncorrelated noise (e.g. ideal scaling) to assess unwanted intervoxel correlation. Later extension introduced the use of radius of decorrelation (RDC), which measures the size of ROI at which statistical independence of the voxel is lost; and adopted quadratic detrending in calculating the CoV without explicit justification. Our study evaluated how detrending choices alter Weisskoff analysis, including RDC, and how they influence their use in routine QC.
Methods
Weisskoff analysis with RDC calculation was performed on gradient‑echo EPI time series acquired on 4 scanners from 3 vendors (two 3T diagnostic and two 1.5T MR-Linac scanners) using site fMRI QC protocol with uniform phantoms. For each dataset, the temporal CoV were computed without detrending, with linear detrending, and with quadratic detrending. Weisskoff analysis plots and RDCs were used to quantify the effects of detrending.
Results
Across platforms and coil configurations, Weisskoff analysis without detrending consistently yielded the largest deviation from the ideal scaling and the smallest RDC, followed by linear, then quadratic detrending, with linear and quadratic often similar. This ordering is consistent with removal of low‑frequency components that reflect slowly varying hardware instabilities, such as gradient heating, thereby increasing the apparent decorrelation radius.
Conclusion
Detrending choice substantially changes RDC and affects the interpretation of temporal stability. Omitting detrending yields a single, easy‑to‑track parameter that encompass all temporal fluctuations, which is useful for longitudinal hardware monitoring, but conflates hardware drifts with other correlated noise sources. Alternatively, polynomial detrending allows a separate drift analysis with the fitting parameters for a more complete QC. Our work suggests that it is prudent to select the same detrending strategy for consistent monitoring and interpretation of fMRI QC.