Early Prediction of Imminent Intrinsically Generated Trips During Gait in People Post-Stroke

<title>Abstract</title> <p> <bold>Purpose:</bold> Falls for people post-stroke often result from an internally generated trip (IGT) while walking. Foot-drop stimulators and bracing can increase foot-clearance during swing, potentially avoiding an IGT, but may be unnecessary for all people post-stroke. Furthermore, stimulating at foot-off may be too late to intervene due to muscle electromechanical delay. As such, we developed an early and targeted IGT predictor to inform timely intervention, only when necessary. <bold>Methods:</bold> First, we collected data from 11 people post-stroke, walking on a treadmill. IGTs and smaller disturbances were grouped together to form an abnormal class, while the majority of gait cycles constituted the normal class. Using a leave-one-participant-out cross-validation, we then tested two methods to resolve this imbalance during a forward sequential feature selection. We then wrapped our two-feature linear model in a counter to enable early IGT prediction. <bold>Results:</bold> Our method predicted 88% of ensuing IGTs with a false alarm rate of 32%, for the average participant. Relative to trip onset, our predictions were made (191±77 ms) significantly earlier (p=0.006) than foot-off (86±39 ms). <bold>Conclusion:</bold> The early and accurate prediction of IGTs, and limited false alarm rate, suggest this method can guide engineering interventions to prevent IGTs in real-time in the future. If successful, such interventions may be used to prevent IGTs in other populations at risk of trip-related falls. </p>