Fig. 5: Results of the longitudinal analysis presented in 6 subplots illustrating different markers over time, covering the clinically assessed rating as well as time series markers measured digitally.
From: Leveraging machine learning for digital gait analysis in ataxia using sensor-free motion capture
The relative change to baseline of (A) the on-site SARA gait score (human baseline, ground truth, (B) the overall, and (C0–C3) the stage-dependent best features are plotted against time since baseline in days. The anatomical miniature (facing the reader) in the bottom left corner of each time series plot (B, C0-C3) indicates from which time series the respective best feature originated. For the stage-dependent analysis, the ataxia patient cohort was divided into subgroups of ataxic gait severity levels according to the baseline SARA gait score of 0 (C0), 1 (C1), 2 (C2), and 3 (C3). The black line represents the mean trajectory, created by a linear interpolation, with the gray shaded area being the standard deviation. The yellow dotted line illustrates the linear regression fit of the given data. Trajectories are color-coded according to the baseline SARA gait score. Pearson’s correlation coefficient and a p-value, using a Wald test, from the linear regression analysis are given. The p-values are corrected for multiple comparisons using Benjamini-Hochberg. Spearman’s rank correlation coefficient (S.Corr.) and R2 are shown to provide complementary views on monotonic trends and explained variance. TS = time series, R = right, L = left. The exact p-values for those TS features showing a significant trend are: B: 0.00014, C0: 0.0374, C1: 0.00357, C2: 1.346 ∗ 10−5, C3: 0.036.
