Fig. 4: Distribution and the cumulative percentage of the averaged FI value in each MDS-UPDRS item scored by the clinician and Model-w-3D-tpi in our dataset.

a Left-hand finger taps; b right-hand finger taps. To facilitate hand parameter comparison between files, we averaged the time-series FI value to depict its characteristics. We utilized kernel density estimation to visualize the distribution within each score group. On the x-axis, you can find the FI values for each score group, while the corresponding number of values is expressed as a density on the y-axis. Notably, the MDS-UPDRS item scores serve as indicators of motor movement severity, exhibiting a strong correlation with the averaged FI value. A higher FI value signifies an individual’s proficient performance in finger taps during recording. Furthermore, the cumulative percentage plot unveils the distinctions among score groups while also highlighting the median and quantile within each group. Remarkably, the cumulative percentage plot pattern for each MDS-UPDRS item estimated by our model closely mirrors that of each MDS-UPDRS item scored by the clinician. Furthermore, the FI value (arbitrary units per second, A.U./s) can be converted to a frequency-peak value (thumb lengths per second, thumb-length/s) using the formula provided in Supplementary Fig. 5. For instance, the median FI value (as depicted in Table 1) for individuals with an MDS-UPDRS item score of 0 for left hand is 0.267 (A.U./s), which corresponds to a converted value of 1.42 (thumb-length/s). Similarly, in the case of individuals with an MDSUPDRS item score of 3 for left hand, the value can be converted from 0.042 (A.U./s) to 0.35 (thumb-length/s).