Cross-Phase Matching in Dynamic Time Warping of Gait Data : The Utility of Phase-Specific Analysis

This study aimed to determine whether cross-phase matching-specifically, alignments that connect the stance and swing phases-occurs when dynamic time warping (DTW) is applied to a full gait cycle, and to quantify its impact by comparison with a phase-specific approach in which cross-phase matching cannot occur. Using 3D motion-capture data from 20 participants, we performed 240 DTW computations across three joints (hip, knee, and ankle). When DTW was applied to the full gait cycle, cross-phase matching occurred in more than 79% of cases for both normal and abnormal gait. Notably, it occurred more frequently in abnormal-gait analyses (>90%), and this difference was statistically significant. No statistically significant differences were observed among joints. Within a gait cycle, the number of cross-phase matching events was fewer than five in 82.27% of cases and five or more in 17.73%. To evaluate the influence of cross-phase matching, we compared changes in optimal warping path (OWP) length and cost sum between the full gait cycle and the phase-specific method. For normal gait, OWP length differed between methods, whereas no significant difference was observed for abnormal gait. In contrast, the cost sum differed significantly between methods for both normal and abnormal gait. These findings indicate that applying DTW to a full gait cycle carries a high risk of cross-phase matching, which can bias OWP length and cost sum and thereby introduce error into study outcomes. Accordingly, a phase-specific approach is recommended to prevent cross-phase matching when applying DTW to gait data.