Recurrence quantification analysis of gait for computer-aided diagnosis of Parkinson's disease.

Introduction: Parkinson's disease (PD) is a typical neurodegenerative disorder characterized by progressive motor impairments. Gait analysis offers a promising avenue for non-invasive PD diagnosis, yet extracting discriminative features from gait signals remains challenging. Methods: This study proposed an intelligent diagnostic framework for PD based on quantitative analysis of recurrence plots derived from plantar pressure gait signals. Gait data were collected from 61 PD patients and 48 healthy subjects using a wearable gait acquisition system. Various recurrence plots representations, including thresholded, non-thresholded, basic, cross, and joint recurrence plots were constructed. From these recurrence plots, traditional recurrence quantification analysis (RQA) features and novel recurrence plot entropy features derived from compressed one-dimensional sequences of non-thresholded recurrence plots were extracted. Results: Statistical analysis revealed that recurrence plot entropy features particularly from cross recurrence plots exhibited superior discriminability. Pressure signals from the heel and toe positions showed the highest specificity. For classification, an integrated feature set combining temporal, pressure, and recurrence plot features achieved the best diagnostic performance using a Cubic Support Vector Machine (CSVM) model, yielding a maximum accuracy of 92.71% in distinguishing PD patients from healthy controls (HC). Discussion: The results demonstrated that the proposed quantitative recurrence plots analysis framework provides a highly effective and automated approach for intelligent PD diagnosis based on gait dynamics.