Figure 6. Pre-operative diffusivity along the optic tract predicts visual recovery.

A support vector regression (SVR, linear kernel) model was trained on the along-tract diffusivity indices, separately for each of the four diffusivity measurements (fractional anisotropy - FA, mean diffusivity - MD, axial diffusivity - AD, and radial diffusivity – RD). The support vectors were trained on 11 hemispheres, and then tested on the 12^th hemisphere. The analysis was jackknifed across the 12 hemispheres, each time leaving one out for test and training on the remaining 11. (A). The graph plots explained variance in visual recovery across all hemispheres in the sample; while support vectors trained on along-tract measurements of radial diffusivity could predict 49% of the variance in the observed data, equivalent analyses with axial diffusivity predicted (a non-significant) 17% of the variance. (B). Null distributions were bootstrapped using permutation tests over randomly shuffled data; histograms plot explained variance (r squared), for permutations over axial and radial diffusivity data. The vertical red line indicates the performance of the model on un-shuffled data (from panel A), and written percentages indicate where model performance fell along the bootstrapped null distribution.