Towards Automated Performance Status Assessment: Temporal Alignment of Motion Skeleton Time Series

Patient Performance Status (PS) is used in cancer medicine to predict prognosis and prescribe treatment. Today, PS assessments rely on assessor’s observation, which is susceptible to biases. A motion tracking system can be used to supplement PS assessments, by recording and analyzing patient’s movement as they perform a standardized mobility task e.g. getting up from office chair to sit on examination table. A temporal alignment of the extracted motion skeleton time series is then needed to enable comparison of corresponding motions in mobility task across recordings. In this paper, we apply existing state-of-the-art temporal alignment algorithms to the extracted time series and evaluate their performance in aligning the keyframes that separate corresponding motions. We then identify key characteristics of these time series that the existing algorithms are not able to exploit correctly: task left-right invariance and vertical-horizontal relative importance. We thus propose Invariant Weighted Dynamic Time Warping (IW-DTW), which takes advantage of these key characteristics. In an evaluation against state-of-the-art algorithms, IW-DTW outperforms them in aligning the keyframes where these key characteristics are present.