Within-Session Reliability of fNIRS in Robot-Assisted Upper-Limb Training

Functional near-infrared spectroscopy (fNIRS) seems opportune for neurofeedback in robot-assisted rehabilitation training due to its noninvasive, less physical restriction, and no electromagnetic disturbance. Previous research has proved the cross-session reliability of fNIRS responses to non-motor tasks (e.g., visual stimuli) and fine-motor tasks (e.g., finger tapping). However, it is still unknown whether fNIRS responses remain reliable 1) in gross-motor tasks, 2) within a training session, and 3) for different training parameters. Hence, this study aimed to investigate the within-session reliability of fNIRS responses to gross-motor tasks for different training parameters. Ten healthy participants were recruited to conduct right elbow extension-flexion in three robot-assisted modes. The Passive mode was fully motor-actuated, while Active1 and Active2 modes involved active engagement with different resistance levels. FNIRS data of three identical runs were used to assess the within-session reliability in terms of the map- (
${R}^{{2}}$
) and cluster-wise (
${R}_{\textit {over}\textit {lap}}$
) spatial reproducibility and the intraclass correlation (ICC) of temporal features. The results revealed good spatial reliability (
${R}^{{2}}$
up to 0.69,
${R}_{\textit {over}\textit {lap}}$
up to 0.68) at the subject level. Besides, the within-session temporal reliabilities of Slope, Max/Min, and Mean were between good and excellent (
$0.60< $
ICC < 0.86). We also found that the within-session reliability was positively correlated with the intensity of the training mode, except for the temporal reliability of HbO in Active2 mode. Overall, our results demonstrated good within-session reliability of fNIRS responses, suggesting fNIRS as reliable neurofeedback for constructing closed-loop robot-assisted rehabilitation systems.