Precision Functional Gradients via Optimal Shrinkage Denoising

Motivation: Resting-state functional MRI gradients are widely used to study functional connections' hierarchies in the neocortex. Due to noise, these gradients are usually analyzed at population level, leaving individual gradients underexplored. Goal(s): We develop a method that produces high-quality individual gradients robust to noise. Approach: We applied block-based optimal shrinkage denoising to rs-fMRI data, then used diffusion embedding to derive functional gradients. Results: Denoising improved temporal signal-to-noise ratio eightfold. Resulting individual gradients are smoother, less noisy, and consistent with established literature for both adult and infant fMRI. Our findings suggest that acquisition time can be reduced by 16 times without losing gradient quality. Impact: Denoising enables high-quality individual gradients, even with temporally undersampled data, allowing precise assessment of individual gradient changes.