Test-retest reliability in fMRI of language: group and task effects

This paper explores how the test-retest reliability is modulated by different groups of participants and experimental tasks. A group of 12 healthy participants and a group of nine stroke patients performed the same language imaging experiment twice, test and retest, on different days. The experiment consists of four conditions, one audio condition and three audiovisual conditions in which the hands are either resting, gesturing, or performing self-adaptive movements. Imaging data were analyzed using multiple linear regression and the results were further used to generate receiver operating characteristic (ROC) curves for each condition for each individual subject. By using area under the curve as a comparison index, we found that stroke patients have less reliability across time than healthy participants, and that when the participants gesture during speech, their imaging data are more reliable than when they are performing hand movements that are not speech-associated. Furthermore, inter-subject variability is less in the gesture task than in any of the other three conditions for healthy participants, but not for stroke patients.