Gabor filter for enhanced recognition of assisted turning events

Nursing facility residents at high risk for pressure ulcers are turned in bed by staff at regular intervals. In an ongoing large, multi-site randomized clinical trial (RCT) the impacts of 2, 3, and 4 hour turning intervals are being studied on pressure ulcer outcomes. In the RCT, it is necessary to objectively confirm the reported times of assisted turning by certified nurse assistants using activity data. Activity peaks in Gaussian-smoothed activity data were unable to confirm all turning events in a supporting pilot study with observer notes of assisted turns. A new technique is presented using one-dimensional Gabor filters that enhances activity peaks due to turning events and diminishes the signal during extended activity. As a result, all assisted turns in the pilot study were detected and confirmed. Moreover, the dependence of detection sensitivity on decision threshold is lowered in this method.