Fig. 5: Classification and prediction of participant’s response via features extraction and statistical learning of the tremor movement. | Nature Communications

Fig. 5: Classification and prediction of participant’s response via features extraction and statistical learning of the tremor movement.

From: Non-invasive suppression of essential tremor via phase-locked disruption of its temporal coherence

Fig. 5

a Exemplary recordings of tremor movement from a participant that showed a reduction in tremor amplitude during phase-locked stimulation relative to sham (i–iii) and one that did not (iv–vi). b Classification accuracy (blue) and F-score (orange) of participants’ response as a function of the number of features. Shown are mean and st.d. values of the tenfold cross-validation. c Most informative features of the class structure. Shown are the 40 top predictive features in (b), clustered according to correlation coefficient and re-ordered according to the clustering; green box, outline of a feature cluster; red square, central feature of a cluster. See Supplementary Table 9 for a list of the features at the cluster’s centre. d Normalised magnitude of exemplary features shown in (c) at the center of the clusters of correlated features. Green, ‘responders’ participants; magenta, ‘non-responders’ participants. See Supplementary Table 9 for description of the features. e Classification accuracy of participants’ response using the 14 most informative features, i.e. the features shown in (c) at the centres of the clusters of correlated features, showing (i) mean classification accuracy ± st.d. vs. number of features, each repeated 100 times with a random selection of features out of the 14 most informative features, and (ii) 2D principal component analysis (PCA) plots of classification using all 14 features. Acc classification accuracy, PC principal component. f Euclidean distance between feature centroids of individual participants and the feature centroids of the responders’ and non-responders’ classes, using the 14 most informative features; *, indicates ‘responders”; green bar, distance to responders class < 0.5 & distance to responders class < distance to non-responders class; magenta bar, distance to responders class > distance to non-responders class. g Same as (f) but for a new cohort of participants, showing distances to the same centroids of responders’ and non-responders’ classes in (f), i.e. of the original participants; grey bar, distance to responders class < 0.5 but distance to responders class > distance to non-responders class. Source data are provided as a Source Data file.

Back to article page