Clinical validation of a rule-based decision tree algorithm for classifying hip movements in people with spinal cord injury.

objectiveTo assess a rule-based decision tree algorithm's performance for classifying and counting specific hip flexion repetitions in able-bodied people and to validate the algorithm's efficacy for people with spinal cord injury (SCI). Alternative placement of the accelerometer was tested. STUDY

designA validation study.

settingSpecialized SCI center in Denmark.

methodsTen able-bodied people and 10 people with SCI were recruited. All participants completed a 15-minute predefined protocol with the following movements: hip flexion in supine 90°, 45° and 20°, hip abduction, pelvic lift, transfer from supine to sitting, sit-to-stand, transfer to a wheelchair, pushed in a wheelchair, Motomed cycling, walking and steps in Nustep fitness trainer. All wore accelerometers on the thigh and a chest-mounted GoPro camera to establish ground truth.

resultsConfusion matrixes showed that able-bodied people's activities and specific hip movements can be classified and the number of repetitions counted with 0.86 accuracy. The algorithm's performance did not change substantially depending on the position of the accelerometer. For people with movement deficits caused by SCI, the accuracy lowered to 0.66 but could be improved to 0.79 for classifying and counting this population's activities/movements overall.

conclusionThe algorithm tested could classify specific hip movements and other activities in the SCI population. This method using a single accelerometer may be applied in clinical trials for people with SCI to objectively assess the change in the number of repetitions over time of hip flexion movements, walking and sit-to-stand activities and to some extent hip abduction and pelvic lift.