Test-retest reliability of fMRI experiments during robot-assisted active and passive stepping

Abstract

Background: Brain activity has been shown to undergo cortical and sub-cortical functional reorganisation over the course of gait rehabilitation in patients suffering from a spinal cord injury or a stroke. These changes however, have not been completely elucidated by neuroimaging to date, mainly due to the scarcity of long-term, follow-up investigations. The magnetic resonance imaging (MRI) compatible stepper MARCOS was specifically developed to enable the investigation of the supraspinal adaptations in paretic patients undergoing gait-rehabilitation in a controlled and repeatable manner. In view of future clinical research, the present study aims at examining the test-retest reliability of functional MRI (fMRI) experiments using MARCOS.

Methods: The effect of repeated active and passive stepping movements on brain activity was investigated in 16 healthy participants from fMRI data collected in two separate imaging sessions six weeks apart. Root mean square errors (RMSE) were calculated for the metrics of motor performance. Regional overlap of brain activation between sessions, as well as an intra-class correlation coefficient (ICC) was computed from the single-subject and group activation maps for five regions of interest (ROI).

Results: Data from eight participants had to be excluded due to excessive head motion. Reliability of motor performance was higher during passive than active movements, as seen in 4.5- to 13-fold lower RMSE for passive movements. In contrast, ICC ranged from 0.48 to 0.72 during passive movements and from 0.77 to 0.85 during active movements. Regional overlap of activations was also higher during active than during passive movements.

Conclusion: These findings imply that an increased variability of motor performance during active movements of healthy participants may be associated with a stable neuronal activation pattern across repeated measurements. In contrast, a stable motor performance during passive movements may be accompanied by a confined reliability of brain activation across repeated measurements.

Fig. 1

The investigational set-up as applied in the study. For the purpose of the present study, the MR-compatible robot MARCOS was mounted on the bench of the MR-scanner. The robot was used to measure and control the delivery of active and passive stepping movements (a). During task execution, the word “MOVE” was presented on the screen and participants conducted stepping movements in the rhythm of the concurrently presented auditory stimulus (metronome). This was followed by the acquisition of the BOLD-signal, while participants fixated on a white cross on the screen. Subsequently, the word “LISTEN” and the metronome were presented concurrently, again followed by the acquisition of the BOLD-signal (b)

Fig. 2

Motor performance and its reliability during passive and active40 stepping. Motor performance at session 1 (t1) and 2 (t2) and root mean squared error (RMSE) of differences between t1 and t2 of the individual participants during passive (left column) and active40 (right column) stepping. a knee amplitude, b stepping frequency and (c) foot force. Rows 1, 3, 5: mean ± one standard deviation at t₁ and t₂. Rows 2, 4, 6: RMSE of differences between t₁ and t₂

Fig. 3

Activation maps during passive and active40 stepping. Top row: Regions of significant BOLD signal increase during passive (a) and active40 (b) stepping at session 1 (t₁) and 2 (t₂), and their overlap. Bottom row: Areas of significantly higher BOLD signal increase at either t₁ or t₂ for passive (c) and active40 (d). Time between t₁ and t₂ ranged between 42 and 48 days. The sections were taken at the z-coordinate indicated at the bottom left of each section, images are displayed in neurological convention (i.e., left is left), p ≤ 0.001, cluster corrected, k = 42 voxels

Fig. 4

Individual test-retest reliability of brain activation in the investigated regions of interest. Reliability of individual activations (ICCsingle) is given for the regions of interest (ROI) cerebellum, M1, S1, S2 and SMA between t1 and t2. Some participants demonstrate consistently higher ICC than others. T-values were extracted from each ROI and reliability was assessed during passive (left) and active 40 (right) stepping. M1 = primary motor cortex, S1 = primary somatosensory cortex, S2 = secondary somatosensory cortex, SMA = supplementary motor area

Fig. 5

Voxel-wise maps of intra-class correlation coefficients for repeated sessions of passive and active40 stepping. Maps of intra-class correlation coefficients (ICCgroup) for repeated sessions of passive (a) and active40 (b) stepping shown on different axial slices (the z-coordinate is indicated at the top of each slice). Bilateral S2 and the paracentral lobule show high ICC in both conditions, while occipital, posterior parietal and prefrontal regions show high ICC as well in active40. Areas with high ICC (red) are hence not necessarily congruent with areas of activation above threshold. Images are displayed in neurological convention (i.e. left is left). The scale on the right indicates the ICC