Freezing-of-Gait detection using temporal, spatial, and physiological features with a support-vector-machine classifier.

Freezing-of-Gait (FoG) is a syndrome of Parkinson's disease defined by episodes when patients show a complete inability to take a step or continue with their locomotion. In order to develop closed-loop therapeutic strategies, including deep brain stimulation, a reliable means of detecting freezing episodes is required. By using wearable accelerometers, freezing episodes can be detected with energy-based algorithms when the ratio of the energy in the freeze band (3 to 8 Hz) to that of the locomotion band (0.5 to 3 Hz) is above a patient-specific threshold. However, due to the great variability in patient activity type, walking style, and freezing pattern, this detection method often does not work. Here we describe a new FoG-detection method that captures temporal, spatial, and physiological features and uses a support-vector-machine to classify freezing episodes. Since our method uses more diverse features, it is able to more robustly detect FoG events. We have shown that when the energy-based method fails (e.g., area under the receiver operator curve is ~0.5), our new method performs well (e.g., area under ROC curve is 0.96).