FIACH: A biophysical model for automatic retrospective noise control in fMRI.

Different noise sources in fMRI acquisition can lead to spurious false positives and reduced sensitivity. We have developed a biophysically-based model (named FIACH: Functional Image Artefact Correction Heuristic) which extends current retrospective noise control methods in fMRI. FIACH can be applied to both General Linear Model (GLM) and resting state functional connectivity MRI (rs-fcMRI) studies. FIACH is a two-step procedure involving the identification and correction of non-physiological large amplitude temporal signal changes and spatial regions of high temporal instability. We have demonstrated its efficacy in a sample of 42 healthy children while performing language tasks that include overt speech with known activations. We demonstrate large improvements in sensitivity when FIACH is compared with current methods of retrospective correction. FIACH reduces the confounding effects of noise and increases the study's power by explaining significant variance that is not contained within the commonly used motion parameters. The method is particularly useful in detecting activations in inferior temporal regions which have proven problematic for fMRI. We have shown greater reproducibility and robustness of fMRI responses using FIACH in the context of task induced motion. In a clinical setting this will translate to increasing the reliability and sensitivity of fMRI used for the identification of language lateralisation and eloquent cortex. FIACH can benefit studies of cognitive development in young children, patient populations and older adults.