JMIR Publications

ABSTRACT

Background:

Expensive optoelectronic systems, considered as the gold standard, require a laboratory environment as well as the attachment of markers and are therefore, rarely used in everyday clinical practice. 2D human pose estimations for clinical purposes, allow kinematic analyses to be carried out via a smartphone, camera-based application. Since clinical specialists are highly dependent on the validity of the information, there is a need to evaluate the accuracy of 2D pose estimation apps.

Objective:

The aim of the study was to investigate the accuracy of the 2D pose estimation of a mobility analysis app (Lindera-v2), based on the PanopticStudio Toolbox as ground truth. The study aimed to assess the differences in joint angles obtained by two-dimensional video-information, generated with the Lindera-v2 algorithm and the ground truth. The results provide an important assessment of the adequacy of the application for clinical use.

Methods:

To evaluate the accuracy of the Lindera-v2 algorithm, 10 video sequences were analyzed. The accuracy was evaluated by comparing a total of 30,000 data pairs for each joint (in total 10 joints), with the angles obtained from the Lindera-v2 algorithm and with those of the PanopticStudio toolbox data set as ground truth. The mean differences (MD) of the angles were calculated for each joint between the estimated values and the ground truth values. Furthermore, the mean absolute error (MAE), root mean squared error (RMSE) and the symmetric mean absolute percent error (sMAPE) of the 2D angles were calculated. The agreement between the two measurement methods was calculated by intraclass correlation ICC(A,2). A cross correlation was calculated for the time series to verify whether there was a temporal shift of the data.

Results:

The MD of Lindera-v2 was in the right hip was the closest to the ground truth, with a mean value difference of -0.05° (SD: 6.06°). The greatest difference to the baseline could be found in the neck with a measurement of -3.07° (SD: 6.43°). The MAE of the angle measurement closest to the baseline, was observed in the pelvis (1.40°; SD: 1.48°), by contrast, the largest mean absolute error was observed in the right shoulder (6.48°; SD: 8.43°). The medians of all acquired joints ranged in difference from 0.19° up to 3.17° by comparison with the ground truth. The ICC values ranged from .951 (95%-CI: 0.914 - 0.969) in the neck to .997 (95%-CI: 0.997 - 0.997) in the left elbow joint. The cross correlation showed that the Lindera-v2 algorithm had no temporal lag.

Conclusions:

The results of the study indicate that a 2D pose estimation by means of a smartphone application can have excellent agreement, compared with a validated ground truth. An assessment of kinematical variables can be performed with the analyzed algorithm and show only minimal deviations, compared with data from a massively multiview system.