Article

Accurate automated event detection is important in increasing the efficiency and
utility of instrumented gait analysis. Published automated event detection
algorithms, however, have had limited testing on pathological populations,
particularly those where force measurements are not available or reliable. In
this study we first postulated robust definitions of gait events that were
subsequently used to compare kinematic based event detection algorithms across
difficult pathologies. We hypothesized that algorithm accuracy would vary by gait
pattern, and that accurate event detection could be accomplished by first
visually classifying the gait pattern, and subsequently choosing the most
appropriate algorithm. Nine published kinematic event detection algorithms were
applied to an existing instrumented pediatric gait database (primarily cerebral
palsy pathologies), that were categorized into 4 visually distinct gait patterns.
More than 750 total events were manually rated and these events were used as a
gold standard for comparison to each algorithm. Results suggested that for foot
strike events, algorithm choice was dependent on whether the foot's motion in
terminal swing was more horizontal or vertical. For horizontal foot motion in
swing, algorithms that used horizontal position, resultant sagittal plane
velocity, or horizontal acceleration signals were most robust; while for vertical
foot motion, resultant sagittal velocity or vertical acceleration excelled. For
toe off events, horizontal position or resultant sagittal plane velocity
performed the best across all groups. We also tuned the resultant sagittal plane
velocity signal to walking speed to create an algorithm that can be used for all
groups and in real time.
CI - Published by Elsevier B.V.

Langue : ANGLAIS