Personalized Prediction of Differences in fMRI Connectivity Estimates Across Preprocessing Pipelines

A ridge regression model trained on 69 ADNI participants predicts how reversing nuisance regression and temporal filtering will change fMRI functional connectivity across 79,800 connections, using only four shared coefficients.

• Why it matters: Preprocessing choices can shift measured connectivity in person-specific ways, complicating longitudinal comparisons for disease progression or treatment response.
• What they did: Using three longitudinal scans per participant, the model learns how nuisance regression and filtering alter FC and combines these with baseline FC to forecast the effect of reversing processing order.
• The result: Leave-one-participant-out testing achieved mean squared error 0.012 versus 0.017 for a population-average predictor, reducing error in 44 of 69 participants and lowering mean error in 10 Midnight Scan Club participants without refitting.

AI summary of the abstract · may contain errors · not medical advice · read the paper.