Fig. 4: Model performance in extracting motion markers of disease progression.

a The four joints with the highest Spearman correlations to UPDRS scores among all 20 joints for each type of extracted spatiotemporal biomarkers (additional information on the remaining joints are shown in Supplementary Fig. 6). The ranges of the p-values of the Spearman correlations for each joint are displayed on the corresponding bars. We selected the largest correlation coefficients between the left and right joints for bilaterally symmetrical joints. b Joint contributions to predict the severity of gait impairment cross all UPDRS scores and each score. The joint contributions were normalized to represent the relative contribution ratios among joints for severity prediction, indicated through points with different sizes and colors. Larger and red points indicate greater normalized contributions, whereas smaller and blue joints denote lesser normalized contributions. A contribution value of 0 means the lowest contribution ratio within the analyzed joints rather than the absence of contribution. c–e The Spearman correlations between the UPDRS scores and the extracted motion markers, along with the significant differences in these markers across three disease severity categories. Spearman correlation coefficients and their p-values are displayed at the bottom of each graph. We used the Kruskal–Wallis test to analyze the statistical significance. p-values of the significance analyses between different groups are presented at the top of each graph.