Work

Preprocessing Determines Whether Fractal Biomarkers Detect Neurodegeneration: A Systematic Evaluation Across Five fMRI Pipelines

Year: 2026
Type: preprint
Abstract: Fractal analysis of resting-state functional MRI (rs-fMRI) has been proposed for detecting altered neural dynamics in Alzheimer’s disease (AD), but the effect of preprocessing on these measurements is poorly constrained. We evaluated 229 ADNI3 participants (156 cognitively normal [CN], 60 mild cognitive impairment [MCI], and 13 AD) processed with five ENIGMA HALFpipe variants. Each scan contained 197 volumes (TR = 3.0 s) and 434 regional time series. The reported CN–AD effect varied 4.5-fold for the detrended fluctuation analysis (DFA) Hurst exponent and 13.1-fold for Sample Entropy; both measures met p < 0.05 only after aCompCor. Under aCompCor, mean H was 0.812 in CN, 0.827 in MCI, and 0.872 in AD (d = −1.40, p = 0.0003), whereas the other four DFA effects were approximately d = −0.31 to −0.40. A variance decomposition attributed 41.5% of total H variance to pipeline, 44.5% to participant, and 0.3% to diagnosis. Whole-brain analysis found higher H in AD across 16 of 18 networks, with 36 cortical regions surviving false-discovery-rate correction. Synthetic fractional Gaussian noise experiments confirmed monotonic DFA estimation at the available scan length. The small AD group makes all disease-effect estimates exploratory. These results show that preprocessing choice can determine whether fractal group differences are detectable and should therefore be treated as part of the biomarker definition.
Language: English
Open Access status: green
OpenAlex ID: W7236687789
Cites: 6
Cited by:
Related to: 10