Estimation of joint angles in the upper limbs using a smartphone.

Abstract

The clinical assessment of spasticity in stroke patients relies on the subjective judgement of specialists. With the aim of objectively quantifying joint mobility, this study validates the BlazePose model—an AI-based model integrated into the MediaPipe framework—to estimate four joint angles of the upper limb. The accuracy of this model is compared with the OptiTrack motion capture system. To this end, an experimental protocol was designed and shoulder abduction and flexion, elbow flexion and wrist flexion were assessed. The results show that the first three angles achieve a correlation coefficient r>= 0.94 with respect to the reference system. Shoulder abduction yields the best result, with a mean absolute error of 8.3º, whilst the elbow shows a systematic bias of +18.9º. The main conclusion of the study is that the clinical viability of this solution is contingent upon specific calibration protocols.