Detection of spatial artifacts on resting state functional magnetic resonance data

Abstract

Brain Functional Magnetic Resonance Image (fMRI) is a widely used non-invasive technique for measuring brain activity and mapping functional regions, but has a complex spatiotemporal structure which complicates the analysis. We present an extension to an existing qualitycontrol (QC) pipeline for resting-state fMRI that automatically detects previously underexplored periodic spatial artefacts. By applying a 3D Fourier transform across each volume and computing inter-slice Pearson correlations over time, we generate summary plots that highlight high-frequency peaks and abnormally elevated correlations, indicative of periodic noise. We integrate these diagnostics in the existing visual and quantitative QC report, allowing reviewers to assign a second periodic-noise PASS/MAYBE/NO-PASS decision based on the presence of periodic noise. In a cohort of 1,178 older adults from the A4 study, our method flagged 42.5% of scans for periodic artifacts that had passed conventional QC. In a classification analysis using a support vector machine with features extracted from the Fourier transform and the spatial correlation analyses against the expert QC labels, we obtained an overall accuracy of 79.5% with a recall for PASS of 92.0%.