Topological Descriptors of Gait Nonlinear Dynamics Toward Freezing-of-Gait Episodes Recognition in Parkinson's Disease

Gait complexity analysis and nonlinear dynamics descriptions were widely used in abnormal state detection in disease studies. Recent development in algebra topology brought novel tools to describe dynamical systems, which inspired us a novel insight toward gait complexity analysis. This work investigates the gait complexity variations in Parkinson's disease patients with a topological nonlinear dynamics analysis approach toward detecting the Freezing-of-Gait episode. In this work, we use 3-dimensional acceleration data from eight subjects with FoG symptoms and extract the FoG features from the accelerometer signals through topological analysis of the reconstructed phase space. Descriptors of Betti curves, persistence landscapes, silhouette landscapes were extracted to form feature vector toward FoG detection tasks, which achieve a sensitivity of 91.93% (94.49%), specificity of 87.61% (93.10%), the accuracy of 89.76% (93.75%), and the receiver operating characteristic curve (AUC) score of 0.956 (0.977) in the generic (person-specific) FoG recognition model. We also found that the proposed approach outperforms the traditional complexity parameters in the FoG recognition task, showing that the topological descriptions are promising in gait analysis. This work is supposed to be the first topological validation work in gait complexity analysis. Our findings are also relevant to risk stratification and continuous monitoring of Parkinson's disease, which bring a promising approach for automated FoG detection for better healthcare.