Gait Event Anomaly Detection and Correction During a Split-Belt Treadmill Task

During instrumented split-belt treadmill tasks, it is challenging to avoid partially stepping on the contralateral belt. If this occurs, accurate detection of gait events from force sensors becomes impossible, as the force data are invalidated. In this paper, we present an algorithm, which automatically detects these invalid force data using an acceleration derivative-based measure. We used this algorithm in combination with the coordinate-based treadmill algorithm to replace the invalidated gait events detected from force sensors with those detected from 3-D markers. The performance of the proposed algorithm was evaluated against the visual examination of data collected from healthy participants in both the same speed and differential speed configurations, using the receiver operator characteristics, the area under the curve, and the Youden index. We found that the area under the curve (AUC) score was above 0.8 in both the same speed and differential speed configurations. Moreover, there was not enough evidence (p > 0.05) to suggest a correlation between walking speed and the performance of the algorithm. We conclude that the algorithm has good to excellent detection and correction performance, which can be useful for research involving analysis of gait with instrumented split-belt treadmills. A MATLAB (MathWorks, Inc., Natick, MA, USA) based implementation of the proposed algorithm and example data files are also presented.