Figure 3

Determining significance of neuro-motor quantifications and comparing with clinical assessments. (A) All labeled and tracked points on the hand. For all presented analyses, frame-to-frame Euclidian displacement was averaged for the five fingertips as this was the minimum number of points needed for sufficient kinematic resolution. (B) Visualization of kinematic (top) and MUA (middle) signals for two cycles of hand clenches, following re-sampling to 1200 Hz and normalization. Overlay of the two signals (bottom) qualitatively demonstrates high similarity. Quantitative comparison of signal similarity using dynamic time-warping (DTW) finds a value of 89.1 a.u. for the sum of sample-to-sample comparisons between extracted kinematic and MUA signals. To determine statistical significance, the null distribution (n = 1000 iterations) percentile corresponding to 89.1 a.u. was calculated. Quantitative comparison to null distribution finds the relationship to be highly significant (1st percentile). (C) Null distribution used for comparison in B. The less the sample-to-sample distance between signals, the more similar signals being compared are. Hence, percentile values closer to 0 represent more similar signals whereas percentile values closer to 100 represent increasingly dissimilar signals. This is demonstrated with the representative iterations (i.e. lower percentile-value comparisons appear more similar on visual inspection). (D) Range of passive movement examples grouped by clinician (clin) assessment: top row, clin + (green); bottom row, clin- (pink). Percentile values in the subplot titles are the scalar output from the automated system analysis described in B & C. Percentile-values are compared against a 5th percentile threshold (i.e. “quant + ” if percentile-value < 5, “quant −” if percentile-value > 5). Quantitative determination is represented by the grey + or − symbols in the subplot titles while clinician assessment is represented by the green + or pink − symbols. (E) Percentile distributions shown separately for clin + (green) and clin − (pink) passive movement epochs. Values closer to 0 indicate comparisons wherein kinematic and MUA signals were quantified as similar by the automated system whereas values closer to 100 indicate dissimilar comparisons. Representative examples in D are shown in relation to the overall distribution of percentile-values and are identified by the notation (clin +/−, quant +/−), as described in D. (F) A receiver operator curve-like analysis for distributions shown in E. The “shuffled” group represents the rank ordered average over 1000 iterations of randomly assigning labels of clin + or clin − to epochs and their corresponding percentile-values. As expected, the shuffled group falls between the clin + and clin − lines. The maximum difference between the latter falls at the 26th percentile. (G) As in D, but for active movements. (H) As in E, but for active movements. (I) As in F, but for active movements. Maximum difference between clin + and clin − falls at 9.5th percentile. When compared to the differences noted in F, both the separation between groups and proportion of clin + epochs ruled in are greater for passive movements, but above chance performance for both movement types. (Note: all components of figure are original work created by AT and designed within Affinity Designer, version 1.10.4).