Does Fractional Anisotropy Predict Motor Imagery Neurofeedback Performance in Healthy Older Adults?

Abstract

Motor imagery neurofeedback training has been proposed as a potential add-on therapy for motor impairment after stroke, but not everyone benefits from it. Previous work has used white matter integrity to predict motor imagery neurofeedback aptitude in healthy young adults. We set out to test this approach with motor imagery neurofeedback that is closer to that used for stroke rehabilitation and in a sample whose age is closer to that of typical stroke patients. Using shrinkage linear discriminant analysis with fractional anisotropy values in 48 white matter regions as predictors, we predicted whether each participant in a sample of 21 healthy older adults (48-77 years old) was a good or a bad performer with 84.8% accuracy. However, the regions used for prediction in our sample differed from those identified previously, and previously suggested regions did not yield significant prediction in our sample. Including demographic and cognitive variables which may correlate with motor imagery neurofeedback performance and white matter structure as candidate predictors revealed an association with age but also led to loss of statistical significance and somewhat poorer prediction accuracy (69.6%). Our results suggest cast doubt on the feasibility of predicting the benefit of motor imagery neurofeedback from fractional anisotropy. At the very least, such predictions should be based on data collected using the same paradigm and with subjects whose characteristics match those of the target case as closely as possible.

Keywords: EEG; MRI; fractional anisotropy; motor imagery; neurofeedback; shrinkage linear discriminant analysis; white matter.

FIGURE 1

Trial structure and relationship between the 2-dimensional neurofeedback display and motor imagery-induced brain activity. The trial structure is illustrated for a right-hand trial. Each trial was initiated with a fixation-cross and after a delay of 3 s a graphic comprising 3 different shades of blue was added. Onset of the graphic indicated the beginning of the task period (duration 5 s). The location of the graphic indicated which hand to use. During the neurofeedback blocks a white circle resembling a ball moved along the horizontal (green arrow) and vertical (orange arrow) axes according to the classifier output magnitudes. Trials were followed by a fixation dot, resulting in an inter-trial interval of 5–9 s. The relationship between the position of the ball and the motor imagery-induced brain activity at the time point of the dashed vertical line is illustrated on an example time course of the event-related desynchronization. The horizontal ball position is determined by the classification of motor imagery contralateral (blue) vs. ipsilateral (red), as illustrated by the green arrow. The vertical ball position is determined by the classification of contralateral baseline (“B”) vs. contralateral motor imagery, as illustrated by the orange arrow. Reproduced, with permission, from Zich et al. (2017a).

FIGURE 2

White matter regions identified as predictive of motor imagery neurofeedback performance by Halder et al. (2013) and in the present study. Regions filled in blue are from the study by Halder et al, where regions 1–5 showed significant correlations with motor imagery neurofeedback performance and regions 6–10 contributed to prediction in a high proportion of cross-validation folds but the correlation with performance was not statistically significant. Regions filled in green are from the present study. Correlation of FA value in the fornix with performance was only significant if age was not included as additional predictor. Total white matter is outlined in white.

FIGURE 3

Correlation between motor imagery neurofeedback performance and mean fractional anisotropy in the left anterior corona radiata (left panel) and mean fractional anisotropy in the column and body of the fornix (right panel). Lines represent regression lines with motor imagery neurofeedback performance (yellow: online; purple: offline) as criterion and mean fractional anisotropy value as predictor. Significant partial correlations between mean fractional anisotropy were found for online performance but not for offline performance – see text for details.