Unveiling Gait Patterns In Parkinson's And Healthy Controls Through Markerless Analysis

Gait abnormalities as lower gait velocity (GV), shorter step length (SL), higher stride (SF) and step frequency (CAD), stride variability (SV) characterise individuals with Parkinson's disease (PD) compared to age-matched controls (C). A markerless, video-based method (i.e., Graal), developed and validated for the quantitative analysis of walking gait in young adults, may be useful to monitor the risk of falls and disease progression in PD. PURPOSE: determine the feasibility and validity of the Graal method vs standard photoelectric sensor system, for the quantitative gait analysis in PD and C. METHODS: 16 participants (8 PD, 64.4 ± 8.6 yrs, 1.6 ± 0.6 H&Y and 8 C, 60.4 ± 7.3 yrs) performed a 5-min treadmill walk at self-selected speed. Average SL, SF, CAD and SV were measured in the final 90 s using a photoelectric sensor system mounted on the treadmill. Concurrently, a posterior video of each participant’s walk was captured by 90-fps webcam and analysed using Graal software to extract the frequency contents of variations of the pixel’s brightness with FFT. We identified the two primary frequencies F1 and F2, calculated SL (based on GV and F2) and an index of gait harmonicity (HI), inversely related to SV. Data were compared between populations and methods by 2-way RM ANOVA and correlation analysis. RESULTS: GV was lower in PD vs C (4.0 ± 0.3 vs 4.5 ± 0.3 km/h, p = 0.02). SL was shorter in PD (57.9 ± 4.5 vs 58.1 ± 4.4 cm, p = 0.05) with no difference between methods. Indexes of movement frequency were not different in PD vs C. In both PD and C, SF and CAD were not different from and highly correlated with F1 (0.959 ± 0.060 vs 0.957 ± 0.059 Hz, p = 0.48, r2 = 0.996; 0.988 ± 0.044 vs 0.983 ± 0.039 Hz, p = 0.41, r2 = 0.961)) and F2 (1.918 ± 0.120 vs 1.917 ± 0.122 Hz, p = 0.49, r2 = 0.995; 1.976 ± 0.089 vs 1.964 ± 0.079 Hz, p = 0.39, r2 = 0.928). SV was higher in PD vs C (3.5 ± 1.0, 2.5 ± 1.1, p = 0.03) with a moderate, negative correlation with HI (r2 = -0.48). CONCLUSION: The markerless, video-based method captures effectively and accurately the distinct gait patterns that part PD from C. Through the identification of the frequency contents of a rhythmic human motion, this approach offers the opportunity for a cost- and time-efficient assessment of human gait. The availability of such a method can facilitate the early detection and frequent monitoring of gait abnormalities. Supported by: Provincia Autonoma di Bolzano-Ufficio Innovazione e Tecnologia