Background: Gait disturbances are an integral part of clinical manifestations of Parkinson's disease (PD), even in the initial stages of the disease. Our goal was to identify the set of spatio-temporal gait parameters that bear the highest relevance for characterizing de novo PD patients.
Methods: Forty patients with de novo PD and forty healthy controls were recorded while walking over an electronic walkway in three different conditions: (1) base walking, (2) walking with an additional motor task, (3) walking with an additional mental task. Both groups were well balanced concerning age and gender. To select a smaller number of relevant parameters, affinity propagation clustering was applied on parameter pairwise correlation. The exemplars were then sorted by importance using the random forest algorithm. Classification accuracy of a support vector machine was tested using the selected parameters and compared to the accuracy of the model using a set of parameters derived from literature.
Results: Final selection of parameters included: stride length and stride length coefficient of variation (CV), stride time and stride time CV, swing time and swing time CV, step time asymmetry, and heel-to-heel base support CV. Classification performed using these parameters showed higher overall accuracy (85%) than classification using the common parameter set containing: stride time, stride length, swing time and double support time, along with their CVs (78%).
Conclusion: In early stages of PD, double support time and its CV appear to be weak indicators of the disease. We instead found step time asymmetry and support base CV to significantly contribute to classification accuracy.
Keywords: Gait parameters; Parkinson’s disease; classification; de novo PD; feature selection.